analysis_yosha.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "33624bb9",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "warning: UnknownSAN in game\n",
      "warning: UnknownSAN in game\n",
      "warning: UnknownSAN in game\n",
      "warning: UnknownSAN in game\n",
      "warning: UnknownSAN in game\n",
      "warning: UnknownSAN in game\n",
      "warning: UnknownSAN in game\n",
      "warning: UnknownSAN in game\n",
      "warning: UnknownSAN in game\n"
     ]
    }
   ],
   "source": [
    "import chess \n",
    "import chess.pgn\n",
    "import glob\n",
    "import pandas as pd\n",
    "import os \n",
    "import re \n",
    "\n",
    "# list each folder of folder \"games\"\n",
    "# get all folders\n",
    "folders = glob.glob(\"games_yosha/*\")\n",
    "\n",
    "# sort folders by date\n",
    "folders.sort(key=os.path.getmtime)\n",
    "\n",
    "df = pd.DataFrame(columns=[\"white\", \"black\", \"white_elo\", \"black_elo\", \"result\", \"nmoves\", \"nstarting_move\", \"pgn_base\", \"temperature\", \"random_engine\", \"has_illegal\", \"illegal_move\", \"folder_name\"])\n",
    "\n",
    "for folder in folders:\n",
    "\n",
    "    # for each folder:\n",
    "    # read PGN file \"game.pgn\"\n",
    "\n",
    "    # check that \"game.pgn\" exists\n",
    "    if not os.path.exists(folder + \"/game.pgn\"):\n",
    "        print(\"No game.pgn in \" + folder) # TODO\n",
    "        continue\n",
    "\n",
    "    with open(folder + \"/game.pgn\") as pgn:\n",
    "        game = chess.pgn.read_game(pgn)\n",
    "\n",
    "    # read metainformation.txt\n",
    "    # get the GPT model and the engine (SF or random)\n",
    "    nmove_value = None\n",
    "    with open(folder + \"/metainformation.txt\") as f:\n",
    "        lines = f.readlines()\n",
    "\n",
    "    # Iterate over each line in the file\n",
    "    for line in lines:\n",
    "        # Check if the line contains the string 'nmove:'\n",
    "        if 'nmove:' in line:\n",
    "            # Split the line at 'nmove:' and take the second part\n",
    "            # Then strip leading and trailing whitespaces and convert to integer\n",
    "            nmove_value = int(line.split('nmove:')[1].strip())\n",
    "            # Print the extracted value\n",
    "            break\n",
    "    \n",
    "    if nmove_value is None:\n",
    "        nmove_value = 1 # default value\n",
    "    \n",
    "    \n",
    "   \n",
    "\n",
    "    # number of moves\n",
    "    game_length = len(list(game.mainline_moves()))\n",
    "\n",
    "\n",
    "\n",
    "\n",
    "    # for each PGN:\n",
    "    # get the GPT model and the engine (SF or random)\n",
    "    white = game.headers[\"White\"]\n",
    "    black = game.headers[\"Black\"]\n",
    "\n",
    "\n",
    "    # get the Elo of the engine (if any)\n",
    "\n",
    "    # get the Elo of the player (if any)\n",
    "    white_elo = game.headers[\"WhiteElo\"]\n",
    "    black_elo = game.headers[\"BlackElo\"]\n",
    "\n",
    "    # get the result (or infer based on checkmates) # special case: no mate, or unifinished game due to wrong move\n",
    "    result = game.headers[\"Result\"]\n",
    "\n",
    "    has_illegal = False\n",
    "    illegal_move = ''\n",
    "    # check that UnknownSAN key is in game\n",
    "    # if not, continue\n",
    "    if 'UnknownSAN' in game.headers:\n",
    "        has_illegal = True\n",
    "        illegal_move = game.headers[\"UnknownSAN\"]\n",
    "        print(\"warning: UnknownSAN in game\")\n",
    "        # continue\n",
    "\n",
    "    with open(folder + \"/metainformation.txt\") as file:\n",
    "        content = file.read()\n",
    "\n",
    "    # find the content between 'temperature:' and ends of line\n",
    "    match_random = re.search(r'random_engine:([\\s\\S]+?)(\\n)', content, re.MULTILINE)\n",
    "    random_engine = None\n",
    "    if match_random:\n",
    "        random_engine = match_random.group(1).strip()\n",
    "        if 'True' in random_engine:\n",
    "            random_engine = True\n",
    "        elif 'False' in random_engine:\n",
    "            random_engine = False\n",
    "        else:   \n",
    "            print(\"random engine value unclear/unknwon\")\n",
    "                \n",
    "    else:\n",
    "        random_engine = False\n",
    "        # print('No random engine found') # default value: False (note: should not happen)\n",
    "\n",
    "    with open(folder + \"/metainformation.txt\") as file:\n",
    "        content = file.read()\n",
    "\n",
    "    # find the content between 'temperature:' and ends of line\n",
    "    match_temperature = re.search(r'temperature:([\\s\\S]+?)(\\n)', content, re.MULTILINE)\n",
    "    temperature = None\n",
    "    if match_temperature:\n",
    "        temperature = match_temperature.group(1).strip()\n",
    "        # print(extracted_content)\n",
    "    else:\n",
    "        temperature = 0.0\n",
    "        # print('No temperature found') # default value: 0\n",
    "\n",
    "    with open(folder + \"/metainformation.txt\") as file:\n",
    "        content = file.read()\n",
    "\n",
    "    # Find the content between 'base_pgn:' and another term at the start of a new line followed by ':'\n",
    "    match = re.search(r'base_pgn:([\\s\\S]+?)(^\\w+:)', content, re.MULTILINE)\n",
    "    pgn_base = ''\n",
    "    if match:\n",
    "        # Extract and clean up the matched content\n",
    "        pgn_base = match.group(1).strip()\n",
    "        # print(extracted_content)\n",
    "    else:\n",
    "        print('No base pgn found')\n",
    "\n",
    "    # put in dfframe\n",
    "    # columns: white, black, white_elo, black_elo, result\n",
    "    # append to dfframe\n",
    "\n",
    "    df = df.append({\"white\": white, \"black\": black, \"white_elo\": white_elo, \"black_elo\": black_elo, \"result\": result, \"nmoves\" : game_length, \"nstarting_move\": nmove_value, \"pgn_base\" : pgn_base, \"temperature\" : temperature, \"random_engine\" : random_engine, \"has_illegal\" : has_illegal, \"illegal_move\" : illegal_move, \"folder_name\" : folder}, ignore_index=True)\n",
    "\n",
    "\n",
    "    # compute stats\n",
    "    # scores in general, per Elo and chess engine, per GPT model\n",
    "    # ability to finish a game (with weaker models)\n",
    "\n",
    "    # first: Elo = 1700, GPT=3.5 instruct\n",
    "    # second: Elo = 1800, GPT=3.5 instruct\n",
    "    # ...\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "3bb30635",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>white</th>\n",
       "      <th>black</th>\n",
       "      <th>white_elo</th>\n",
       "      <th>black_elo</th>\n",
       "      <th>result</th>\n",
       "      <th>nmoves</th>\n",
       "      <th>nstarting_move</th>\n",
       "      <th>pgn_base</th>\n",
       "      <th>temperature</th>\n",
       "      <th>random_engine</th>\n",
       "      <th>has_illegal</th>\n",
       "      <th>illegal_move</th>\n",
       "      <th>folder_name</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>Stockfish</td>\n",
       "      <td>gpt-3.5-turbo-instruct</td>\n",
       "      <td>1785</td>\n",
       "      <td>?</td>\n",
       "      <td>1-0</td>\n",
       "      <td>141</td>\n",
       "      <td>1</td>\n",
       "      <td>[White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td></td>\n",
       "      <td>games_yosha/gameac41f898-b3ca-4b96-90fb-19246e...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>Stockfish</td>\n",
       "      <td>gpt-3.5-turbo-instruct</td>\n",
       "      <td>1785</td>\n",
       "      <td>?</td>\n",
       "      <td>0-1</td>\n",
       "      <td>58</td>\n",
       "      <td>1</td>\n",
       "      <td>[White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td></td>\n",
       "      <td>games_yosha/gamecc87b77a-986f-4846-9830-095662...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>Stockfish</td>\n",
       "      <td>gpt-3.5-turbo-instruct</td>\n",
       "      <td>1785</td>\n",
       "      <td>?</td>\n",
       "      <td>1-0</td>\n",
       "      <td>53</td>\n",
       "      <td>1</td>\n",
       "      <td>[White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td></td>\n",
       "      <td>games_yosha/gameb0a3fbe2-63a6-4fd4-8a69-605854...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>Stockfish</td>\n",
       "      <td>gpt-3.5-turbo-instruct</td>\n",
       "      <td>1785</td>\n",
       "      <td>?</td>\n",
       "      <td>1-0</td>\n",
       "      <td>75</td>\n",
       "      <td>1</td>\n",
       "      <td>[White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td></td>\n",
       "      <td>games_yosha/gamedc31f43d-e0e8-4301-b73f-aed98e...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>Stockfish</td>\n",
       "      <td>gpt-3.5-turbo-instruct</td>\n",
       "      <td>1785</td>\n",
       "      <td>?</td>\n",
       "      <td>*</td>\n",
       "      <td>91</td>\n",
       "      <td>1</td>\n",
       "      <td>[White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>False</td>\n",
       "      <td>True</td>\n",
       "      <td>0-1</td>\n",
       "      <td>games_yosha/gameac4eefd4-b0a5-4289-84e9-7b2f93...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>76</th>\n",
       "      <td>Stockfish</td>\n",
       "      <td>gpt-3.5-turbo-instruct</td>\n",
       "      <td>1954</td>\n",
       "      <td>?</td>\n",
       "      <td>1-0</td>\n",
       "      <td>115</td>\n",
       "      <td>1</td>\n",
       "      <td>[White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td></td>\n",
       "      <td>games_yosha/gamedc232246-8d2e-42b4-bef2-0cc18a...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>77</th>\n",
       "      <td>Stockfish</td>\n",
       "      <td>gpt-3.5-turbo-instruct</td>\n",
       "      <td>1954</td>\n",
       "      <td>?</td>\n",
       "      <td>0-1</td>\n",
       "      <td>88</td>\n",
       "      <td>1</td>\n",
       "      <td>[White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td></td>\n",
       "      <td>games_yosha/gamea8b0b6a6-ad94-45bd-8170-522304...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>78</th>\n",
       "      <td>Stockfish</td>\n",
       "      <td>gpt-3.5-turbo-instruct</td>\n",
       "      <td>1954</td>\n",
       "      <td>?</td>\n",
       "      <td>0-1</td>\n",
       "      <td>134</td>\n",
       "      <td>1</td>\n",
       "      <td>[White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td></td>\n",
       "      <td>games_yosha/game8fdf34d5-b73e-4812-8789-d11428...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>79</th>\n",
       "      <td>Stockfish</td>\n",
       "      <td>gpt-3.5-turbo-instruct</td>\n",
       "      <td>1954</td>\n",
       "      <td>?</td>\n",
       "      <td>1/2-1/2</td>\n",
       "      <td>161</td>\n",
       "      <td>1</td>\n",
       "      <td>[White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td></td>\n",
       "      <td>games_yosha/gamea541823d-a8bb-45fa-af00-adfafc...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>80</th>\n",
       "      <td>Stockfish</td>\n",
       "      <td>gpt-3.5-turbo-instruct</td>\n",
       "      <td>1954</td>\n",
       "      <td>?</td>\n",
       "      <td>1-0</td>\n",
       "      <td>51</td>\n",
       "      <td>1</td>\n",
       "      <td>[White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td></td>\n",
       "      <td>games_yosha/game251270b2-0667-44e8-b49b-384e16...</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>81 rows × 13 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "        white                   black white_elo black_elo   result nmoves  \\\n",
       "0   Stockfish  gpt-3.5-turbo-instruct      1785         ?      1-0    141   \n",
       "1   Stockfish  gpt-3.5-turbo-instruct      1785         ?      0-1     58   \n",
       "2   Stockfish  gpt-3.5-turbo-instruct      1785         ?      1-0     53   \n",
       "3   Stockfish  gpt-3.5-turbo-instruct      1785         ?      1-0     75   \n",
       "4   Stockfish  gpt-3.5-turbo-instruct      1785         ?        *     91   \n",
       "..        ...                     ...       ...       ...      ...    ...   \n",
       "76  Stockfish  gpt-3.5-turbo-instruct      1954         ?      1-0    115   \n",
       "77  Stockfish  gpt-3.5-turbo-instruct      1954         ?      0-1     88   \n",
       "78  Stockfish  gpt-3.5-turbo-instruct      1954         ?      0-1    134   \n",
       "79  Stockfish  gpt-3.5-turbo-instruct      1954         ?  1/2-1/2    161   \n",
       "80  Stockfish  gpt-3.5-turbo-instruct      1954         ?      1-0     51   \n",
       "\n",
       "   nstarting_move                                           pgn_base  \\\n",
       "0               1  [White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...   \n",
       "1               1  [White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...   \n",
       "2               1  [White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...   \n",
       "3               1  [White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...   \n",
       "4               1  [White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...   \n",
       "..            ...                                                ...   \n",
       "76              1  [White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...   \n",
       "77              1  [White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...   \n",
       "78              1  [White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...   \n",
       "79              1  [White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...   \n",
       "80              1  [White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...   \n",
       "\n",
       "   temperature random_engine has_illegal illegal_move  \\\n",
       "0          0.0         False       False                \n",
       "1          0.0         False       False                \n",
       "2          0.0         False       False                \n",
       "3          0.0         False       False                \n",
       "4          0.0         False        True          0-1   \n",
       "..         ...           ...         ...          ...   \n",
       "76         0.0         False       False                \n",
       "77         0.0         False       False                \n",
       "78         0.0         False       False                \n",
       "79         0.0         False       False                \n",
       "80         0.0         False       False                \n",
       "\n",
       "                                          folder_name  \n",
       "0   games_yosha/gameac41f898-b3ca-4b96-90fb-19246e...  \n",
       "1   games_yosha/gamecc87b77a-986f-4846-9830-095662...  \n",
       "2   games_yosha/gameb0a3fbe2-63a6-4fd4-8a69-605854...  \n",
       "3   games_yosha/gamedc31f43d-e0e8-4301-b73f-aed98e...  \n",
       "4   games_yosha/gameac4eefd4-b0a5-4289-84e9-7b2f93...  \n",
       "..                                                ...  \n",
       "76  games_yosha/gamedc232246-8d2e-42b4-bef2-0cc18a...  \n",
       "77  games_yosha/gamea8b0b6a6-ad94-45bd-8170-522304...  \n",
       "78  games_yosha/game8fdf34d5-b73e-4812-8789-d11428...  \n",
       "79  games_yosha/gamea541823d-a8bb-45fa-af00-adfafc...  \n",
       "80  games_yosha/game251270b2-0667-44e8-b49b-384e16...  \n",
       "\n",
       "[81 rows x 13 columns]"
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "0ee48e12",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0\n"
     ]
    }
   ],
   "source": [
    "import re\n",
    "\n",
    "# Open and read the content of metainformation.txt\n",
    "with open(\"games/gamea7e73056-951a-417d-b671-0a60fba939f7/\" + 'metainformation.txt', 'r') as file:\n",
    "    content = file.read()\n",
    "\n",
    "# Find the content between 'base_pgn:' and another term at the start of a new line followed by ':'\n",
    "match = re.search(r'base_pgn:([\\s\\S]+?)(^\\w+:)', content, re.MULTILINE)\n",
    "match = re.search(r'temperature:([\\s\\S]+?)(\\n)', content, re.MULTILINE)\n",
    "\n",
    "if match:\n",
    "    # Extract and clean up the matched content\n",
    "    extracted_content = match.group(1).strip()\n",
    "    print(extracted_content)\n",
    "else:\n",
    "    print('No match found')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "fb1b28bf",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "81 games compiled in the database\n"
     ]
    }
   ],
   "source": [
    "print(len(df), \"games compiled in the database\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "46e66f75",
   "metadata": {},
   "outputs": [],
   "source": [
    "import chess\n",
    "import chess.pgn\n",
    "import io \n",
    "\n",
    "\n",
    "# for column \"pgn_base\", I'd like to transform the string into a list of moves\n",
    "\n",
    "def pgn_to_list(pgn):\n",
    "    g = chess.pgn.read_game(io.StringIO(pgn))\n",
    "    g.mainline_moves()\n",
    "    return str(g.mainline_moves())\n",
    "\n",
    "def is_base_prompt(pgn):\n",
    "    g = chess.pgn.read_game(io.StringIO(pgn))\n",
    "    # [Event \"FIDE World Championship Match 2024\"]\\n[Site \"Los Angeles, USA\"]\n",
    "    return g.headers[\"Event\"] == \"FIDE World Championship Match 2024\" and g.headers[\"Site\"] == \"Los Angeles, USA\"\n",
    "\n",
    "def has_illegal_moves(pgn):\n",
    "    # exist g.headers[\"UnknownSAN\"] \n",
    "    g = chess.pgn.read_game(io.StringIO(pgn))\n",
    "    # key in array\n",
    "    return \"UnknownSAN\" in g.headers\n",
    "\n",
    "df[\"pgn_base_moves\"] = df[\"pgn_base\"].apply(pgn_to_list) # extract only moves\n",
    "df[\"base_pgn_prompt\"] = df[\"pgn_base\"].apply(is_base_prompt) # extract only prompt\n",
    "# df.sort_values(by=['nstarting_move'])\n",
    "df['temperature'] = pd.to_numeric(df['temperature'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "3118f825",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      " * gpt-3.5-turbo-instruct : 81 games, among 0 with white piece and 81 with black pieces\n",
      " * gpt-4 : 0 games, among 0 with white piece and 0 with black pieces\n",
      " * gpt-3.5-turbo : 0 games, among 0 with white piece and 0 with black pieces\n",
      " * text-davinci-003 : 0 games, among 0 with white piece and 0 with black pieces\n"
     ]
    }
   ],
   "source": [
    "# Frequencies based on piece colors\n",
    "players = ['gpt-3.5-turbo-instruct', 'gpt-4', 'gpt-3.5-turbo', 'text-davinci-003']\n",
    "color_breakdown = {}\n",
    "for player in players:\n",
    "    white_count = df[df['white'] == player].shape[0]\n",
    "    black_count = df[df['black'] == player].shape[0]\n",
    "    color_breakdown[player] = {'white': white_count, 'black': black_count}\n",
    "\n",
    "for player in players:\n",
    "    nwhite = color_breakdown[player]['white']\n",
    "    nblack = color_breakdown[player]['black']\n",
    "    print(\" *\", player, \":\", nwhite+nblack, \"games, among\", nwhite, \"with white piece and\", nblack, \"with black pieces\")\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "1c098cd6",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>white</th>\n",
       "      <th>black</th>\n",
       "      <th>white_elo</th>\n",
       "      <th>black_elo</th>\n",
       "      <th>result</th>\n",
       "      <th>nmoves</th>\n",
       "      <th>nstarting_move</th>\n",
       "      <th>pgn_base</th>\n",
       "      <th>temperature</th>\n",
       "      <th>random_engine</th>\n",
       "      <th>has_illegal</th>\n",
       "      <th>illegal_move</th>\n",
       "      <th>folder_name</th>\n",
       "      <th>pgn_base_moves</th>\n",
       "      <th>base_pgn_prompt</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>Stockfish</td>\n",
       "      <td>gpt-3.5-turbo-instruct</td>\n",
       "      <td>1785</td>\n",
       "      <td>?</td>\n",
       "      <td>1-0</td>\n",
       "      <td>141</td>\n",
       "      <td>1</td>\n",
       "      <td>[White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td></td>\n",
       "      <td>games_yosha/gameac41f898-b3ca-4b96-90fb-19246e...</td>\n",
       "      <td></td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>Stockfish</td>\n",
       "      <td>gpt-3.5-turbo-instruct</td>\n",
       "      <td>1785</td>\n",
       "      <td>?</td>\n",
       "      <td>0-1</td>\n",
       "      <td>58</td>\n",
       "      <td>1</td>\n",
       "      <td>[White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td></td>\n",
       "      <td>games_yosha/gamecc87b77a-986f-4846-9830-095662...</td>\n",
       "      <td></td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>Stockfish</td>\n",
       "      <td>gpt-3.5-turbo-instruct</td>\n",
       "      <td>1785</td>\n",
       "      <td>?</td>\n",
       "      <td>1-0</td>\n",
       "      <td>53</td>\n",
       "      <td>1</td>\n",
       "      <td>[White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td></td>\n",
       "      <td>games_yosha/gameb0a3fbe2-63a6-4fd4-8a69-605854...</td>\n",
       "      <td></td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>Stockfish</td>\n",
       "      <td>gpt-3.5-turbo-instruct</td>\n",
       "      <td>1785</td>\n",
       "      <td>?</td>\n",
       "      <td>1-0</td>\n",
       "      <td>75</td>\n",
       "      <td>1</td>\n",
       "      <td>[White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td></td>\n",
       "      <td>games_yosha/gamedc31f43d-e0e8-4301-b73f-aed98e...</td>\n",
       "      <td></td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>Stockfish</td>\n",
       "      <td>gpt-3.5-turbo-instruct</td>\n",
       "      <td>1785</td>\n",
       "      <td>?</td>\n",
       "      <td>*</td>\n",
       "      <td>91</td>\n",
       "      <td>1</td>\n",
       "      <td>[White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>False</td>\n",
       "      <td>True</td>\n",
       "      <td>0-1</td>\n",
       "      <td>games_yosha/gameac4eefd4-b0a5-4289-84e9-7b2f93...</td>\n",
       "      <td></td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "       white                   black white_elo black_elo result nmoves  \\\n",
       "0  Stockfish  gpt-3.5-turbo-instruct      1785         ?    1-0    141   \n",
       "1  Stockfish  gpt-3.5-turbo-instruct      1785         ?    0-1     58   \n",
       "2  Stockfish  gpt-3.5-turbo-instruct      1785         ?    1-0     53   \n",
       "3  Stockfish  gpt-3.5-turbo-instruct      1785         ?    1-0     75   \n",
       "4  Stockfish  gpt-3.5-turbo-instruct      1785         ?      *     91   \n",
       "\n",
       "  nstarting_move                                           pgn_base  \\\n",
       "0              1  [White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...   \n",
       "1              1  [White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...   \n",
       "2              1  [White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...   \n",
       "3              1  [White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...   \n",
       "4              1  [White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...   \n",
       "\n",
       "   temperature random_engine has_illegal illegal_move  \\\n",
       "0          0.0         False       False                \n",
       "1          0.0         False       False                \n",
       "2          0.0         False       False                \n",
       "3          0.0         False       False                \n",
       "4          0.0         False        True          0-1   \n",
       "\n",
       "                                         folder_name pgn_base_moves  \\\n",
       "0  games_yosha/gameac41f898-b3ca-4b96-90fb-19246e...                  \n",
       "1  games_yosha/gamecc87b77a-986f-4846-9830-095662...                  \n",
       "2  games_yosha/gameb0a3fbe2-63a6-4fd4-8a69-605854...                  \n",
       "3  games_yosha/gamedc31f43d-e0e8-4301-b73f-aed98e...                  \n",
       "4  games_yosha/gameac4eefd4-b0a5-4289-84e9-7b2f93...                  \n",
       "\n",
       "   base_pgn_prompt  \n",
       "0            False  \n",
       "1            False  \n",
       "2            False  \n",
       "3            False  \n",
       "4            False  "
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "27a680b6",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "13b053b8",
   "metadata": {},
   "outputs": [],
   "source": [
    "df.to_csv(\"games_db_yosha.csv\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "28c35c6a",
   "metadata": {},
   "outputs": [],
   "source": [
    "def compute_score(df, model_name='gpt-3.5-turbo-instruct', percentage=True):\n",
    "    # Count the number of wins, losses, and draws for gpt-3.5-turbo-instruct\n",
    "    wins_as_white = df[(df['white'] == model_name) & (df['result'] == '1-0')].shape[0]\n",
    "    wins_as_black = df[(df['black'] == model_name) & (df['result'] == '0-1')].shape[0]\n",
    "    losses_as_white = df[(df['white'] == model_name) & (df['result'] == '0-1')].shape[0]\n",
    "    losses_as_black = df[(df['black'] == model_name) & (df['result'] == '1-0')].shape[0]\n",
    "    draws_as_white = df[(df['white'] == model_name) & (df['result'] == '1/2-1/2')].shape[0]\n",
    "    draws_as_black = df[(df['black'] == model_name) & (df['result'] == '1/2-1/2')].shape[0]\n",
    "\n",
    "    # Calculate total wins, losses, and draws\n",
    "    total_wins = wins_as_white + wins_as_black\n",
    "    total_losses = losses_as_white + losses_as_black\n",
    "    total_draws = draws_as_white + draws_as_black\n",
    "\n",
    "    if percentage:\n",
    "        if total_wins + total_losses + total_draws == 0:\n",
    "            return 0.0\n",
    "        return (total_wins + (total_draws * 0.5)) / (total_wins + total_losses + total_draws)\n",
    "    else:\n",
    "        return (total_wins + (total_draws * 0.5), total_wins + total_losses + total_draws)\n"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "id": "d668c8a9",
   "metadata": {},
   "source": [
    "text-davinci-003 (completion)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "c08a9942",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0 games with davinci\n",
      "with white pieces 0\n",
      "with black pieces 0\n",
      "0 games against random chess engine\n",
      "0 games with illegal move\n",
      "score 0.0\n",
      "the only game with no illegal move is\n",
      "The longest game was nan moves\n",
      "The shortest game was nan moves\n",
      "The average game length was nan moves\n",
      "The median game length was nan moves\n"
     ]
    }
   ],
   "source": [
    "df_davinci = df.query(\"white == 'text-davinci-003' or black == 'text-davinci-003'\")\n",
    "print(len(df_davinci), \"games with davinci\")\n",
    "print(\"with white pieces\", len(df.query(\"white == 'text-davinci-003'\")))\n",
    "print(\"with black pieces\", len(df.query(\"black == 'text-davinci-003'\")))\n",
    "\n",
    "print(len(df_davinci.query(\"white == 'RANDOM chess engine' or black == 'RANDOM chess engine'\")), \"games against random chess engine\")\n",
    "# print(compute_score(df_davinci.query(\"white == 'RANDOM chess engine' or black == 'RANDOM chess engine'\"), \"text-davinci-003\"))\n",
    "\n",
    "\n",
    "print(len(df_davinci.query(\"has_illegal == True\")), \"games with illegal move\")\n",
    "print(\"score\", compute_score(df_davinci, \"text-davinci-003\"))\n",
    "\n",
    "print(\"the only game with no illegal move is\")\n",
    "df_davinci.query(\"has_illegal == False\")\n",
    "\n",
    "\n",
    "# box plot of df_davinci['nmoves']\n",
    "print(\"The longest game was\", df_davinci['nmoves'].max()/2, \"moves\")\n",
    "print(\"The shortest game was\", df_davinci['nmoves'].min()/2, \"moves\")\n",
    "print(\"The average game length was\", df_davinci['nmoves'].mean()/2, \"moves\")\n",
    "print(\"The median game length was\", df_davinci['nmoves'].median()/2, \"moves\")\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "eed048b0",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "f965e556",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>white</th>\n",
       "      <th>black</th>\n",
       "      <th>white_elo</th>\n",
       "      <th>black_elo</th>\n",
       "      <th>result</th>\n",
       "      <th>nmoves</th>\n",
       "      <th>nstarting_move</th>\n",
       "      <th>pgn_base</th>\n",
       "      <th>temperature</th>\n",
       "      <th>random_engine</th>\n",
       "      <th>has_illegal</th>\n",
       "      <th>illegal_move</th>\n",
       "      <th>folder_name</th>\n",
       "      <th>pgn_base_moves</th>\n",
       "      <th>base_pgn_prompt</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>Stockfish</td>\n",
       "      <td>gpt-3.5-turbo-instruct</td>\n",
       "      <td>1785</td>\n",
       "      <td>?</td>\n",
       "      <td>1-0</td>\n",
       "      <td>141</td>\n",
       "      <td>1</td>\n",
       "      <td>[White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td></td>\n",
       "      <td>games_yosha/gameac41f898-b3ca-4b96-90fb-19246e...</td>\n",
       "      <td></td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>57</th>\n",
       "      <td>Stockfish</td>\n",
       "      <td>gpt-3.5-turbo-instruct</td>\n",
       "      <td>1871</td>\n",
       "      <td>?</td>\n",
       "      <td>0-1</td>\n",
       "      <td>66</td>\n",
       "      <td>1</td>\n",
       "      <td>[White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td></td>\n",
       "      <td>games_yosha/game759073af-1266-4266-8072-797742...</td>\n",
       "      <td></td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>56</th>\n",
       "      <td>Stockfish</td>\n",
       "      <td>gpt-3.5-turbo-instruct</td>\n",
       "      <td>1871</td>\n",
       "      <td>?</td>\n",
       "      <td>1-0</td>\n",
       "      <td>133</td>\n",
       "      <td>1</td>\n",
       "      <td>[White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td></td>\n",
       "      <td>games_yosha/gamef302ae5e-faaf-4ef3-a3b2-36e22f...</td>\n",
       "      <td></td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>55</th>\n",
       "      <td>Stockfish</td>\n",
       "      <td>gpt-3.5-turbo-instruct</td>\n",
       "      <td>1871</td>\n",
       "      <td>?</td>\n",
       "      <td>0-1</td>\n",
       "      <td>172</td>\n",
       "      <td>1</td>\n",
       "      <td>[White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td></td>\n",
       "      <td>games_yosha/game39187607-e0f5-4521-928e-e46938...</td>\n",
       "      <td></td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>54</th>\n",
       "      <td>Stockfish</td>\n",
       "      <td>gpt-3.5-turbo-instruct</td>\n",
       "      <td>1871</td>\n",
       "      <td>?</td>\n",
       "      <td>0-1</td>\n",
       "      <td>78</td>\n",
       "      <td>1</td>\n",
       "      <td>[White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td></td>\n",
       "      <td>games_yosha/game6592812f-c8c1-4eca-a9aa-2e46f3...</td>\n",
       "      <td></td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>23</th>\n",
       "      <td>Stockfish</td>\n",
       "      <td>gpt-3.5-turbo-instruct</td>\n",
       "      <td>1694</td>\n",
       "      <td>?</td>\n",
       "      <td>0-1</td>\n",
       "      <td>80</td>\n",
       "      <td>1</td>\n",
       "      <td>[White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td></td>\n",
       "      <td>games_yosha/game76feac0f-8e55-49a5-8570-3b92e4...</td>\n",
       "      <td></td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>22</th>\n",
       "      <td>Stockfish</td>\n",
       "      <td>gpt-3.5-turbo-instruct</td>\n",
       "      <td>1694</td>\n",
       "      <td>?</td>\n",
       "      <td>1-0</td>\n",
       "      <td>83</td>\n",
       "      <td>1</td>\n",
       "      <td>[White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td></td>\n",
       "      <td>games_yosha/gamec6fe0ff8-f7d1-4d14-8861-5d9bfe...</td>\n",
       "      <td></td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>21</th>\n",
       "      <td>Stockfish</td>\n",
       "      <td>gpt-3.5-turbo-instruct</td>\n",
       "      <td>1694</td>\n",
       "      <td>?</td>\n",
       "      <td>1-0</td>\n",
       "      <td>49</td>\n",
       "      <td>1</td>\n",
       "      <td>[White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td></td>\n",
       "      <td>games_yosha/game59b1fb0d-b0e7-40e7-9315-bc2bd0...</td>\n",
       "      <td></td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>39</th>\n",
       "      <td>Stockfish</td>\n",
       "      <td>gpt-3.5-turbo-instruct</td>\n",
       "      <td>1785</td>\n",
       "      <td>?</td>\n",
       "      <td>0-1</td>\n",
       "      <td>144</td>\n",
       "      <td>1</td>\n",
       "      <td>[White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td></td>\n",
       "      <td>games_yosha/game50deb62b-ade0-40e7-b906-235e00...</td>\n",
       "      <td></td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>80</th>\n",
       "      <td>Stockfish</td>\n",
       "      <td>gpt-3.5-turbo-instruct</td>\n",
       "      <td>1954</td>\n",
       "      <td>?</td>\n",
       "      <td>1-0</td>\n",
       "      <td>51</td>\n",
       "      <td>1</td>\n",
       "      <td>[White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td></td>\n",
       "      <td>games_yosha/game251270b2-0667-44e8-b49b-384e16...</td>\n",
       "      <td></td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>81 rows × 15 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "        white                   black white_elo black_elo result nmoves  \\\n",
       "0   Stockfish  gpt-3.5-turbo-instruct      1785         ?    1-0    141   \n",
       "57  Stockfish  gpt-3.5-turbo-instruct      1871         ?    0-1     66   \n",
       "56  Stockfish  gpt-3.5-turbo-instruct      1871         ?    1-0    133   \n",
       "55  Stockfish  gpt-3.5-turbo-instruct      1871         ?    0-1    172   \n",
       "54  Stockfish  gpt-3.5-turbo-instruct      1871         ?    0-1     78   \n",
       "..        ...                     ...       ...       ...    ...    ...   \n",
       "23  Stockfish  gpt-3.5-turbo-instruct      1694         ?    0-1     80   \n",
       "22  Stockfish  gpt-3.5-turbo-instruct      1694         ?    1-0     83   \n",
       "21  Stockfish  gpt-3.5-turbo-instruct      1694         ?    1-0     49   \n",
       "39  Stockfish  gpt-3.5-turbo-instruct      1785         ?    0-1    144   \n",
       "80  Stockfish  gpt-3.5-turbo-instruct      1954         ?    1-0     51   \n",
       "\n",
       "   nstarting_move                                           pgn_base  \\\n",
       "0               1  [White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...   \n",
       "57              1  [White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...   \n",
       "56              1  [White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...   \n",
       "55              1  [White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...   \n",
       "54              1  [White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...   \n",
       "..            ...                                                ...   \n",
       "23              1  [White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...   \n",
       "22              1  [White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...   \n",
       "21              1  [White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...   \n",
       "39              1  [White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...   \n",
       "80              1  [White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...   \n",
       "\n",
       "    temperature random_engine has_illegal illegal_move  \\\n",
       "0           0.0         False       False                \n",
       "57          0.0         False       False                \n",
       "56          0.0         False       False                \n",
       "55          0.0         False       False                \n",
       "54          0.0         False       False                \n",
       "..          ...           ...         ...          ...   \n",
       "23          0.0         False       False                \n",
       "22          0.0         False       False                \n",
       "21          0.0         False       False                \n",
       "39          0.0         False       False                \n",
       "80          0.0         False       False                \n",
       "\n",
       "                                          folder_name pgn_base_moves  \\\n",
       "0   games_yosha/gameac41f898-b3ca-4b96-90fb-19246e...                  \n",
       "57  games_yosha/game759073af-1266-4266-8072-797742...                  \n",
       "56  games_yosha/gamef302ae5e-faaf-4ef3-a3b2-36e22f...                  \n",
       "55  games_yosha/game39187607-e0f5-4521-928e-e46938...                  \n",
       "54  games_yosha/game6592812f-c8c1-4eca-a9aa-2e46f3...                  \n",
       "..                                                ...            ...   \n",
       "23  games_yosha/game76feac0f-8e55-49a5-8570-3b92e4...                  \n",
       "22  games_yosha/gamec6fe0ff8-f7d1-4d14-8861-5d9bfe...                  \n",
       "21  games_yosha/game59b1fb0d-b0e7-40e7-9315-bc2bd0...                  \n",
       "39  games_yosha/game50deb62b-ade0-40e7-b906-235e00...                  \n",
       "80  games_yosha/game251270b2-0667-44e8-b49b-384e16...                  \n",
       "\n",
       "    base_pgn_prompt  \n",
       "0             False  \n",
       "57            False  \n",
       "56            False  \n",
       "55            False  \n",
       "54            False  \n",
       "..              ...  \n",
       "23            False  \n",
       "22            False  \n",
       "21            False  \n",
       "39            False  \n",
       "80            False  \n",
       "\n",
       "[81 rows x 15 columns]"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df.sort_values(by=['pgn_base'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "bd2edd56",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>white</th>\n",
       "      <th>black</th>\n",
       "      <th>white_elo</th>\n",
       "      <th>black_elo</th>\n",
       "      <th>result</th>\n",
       "      <th>nmoves</th>\n",
       "      <th>nstarting_move</th>\n",
       "      <th>pgn_base</th>\n",
       "      <th>temperature</th>\n",
       "      <th>random_engine</th>\n",
       "      <th>has_illegal</th>\n",
       "      <th>illegal_move</th>\n",
       "      <th>folder_name</th>\n",
       "      <th>pgn_base_moves</th>\n",
       "      <th>base_pgn_prompt</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>14</th>\n",
       "      <td>Stockfish</td>\n",
       "      <td>gpt-3.5-turbo-instruct</td>\n",
       "      <td>1694</td>\n",
       "      <td>?</td>\n",
       "      <td>1/2-1/2</td>\n",
       "      <td>138</td>\n",
       "      <td>1</td>\n",
       "      <td>[White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td></td>\n",
       "      <td>games_yosha/gamea4187bf2-f3e0-44b2-ab5c-91aedf...</td>\n",
       "      <td></td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>27</th>\n",
       "      <td>Stockfish</td>\n",
       "      <td>gpt-3.5-turbo-instruct</td>\n",
       "      <td>1694</td>\n",
       "      <td>?</td>\n",
       "      <td>1-0</td>\n",
       "      <td>69</td>\n",
       "      <td>1</td>\n",
       "      <td>[White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td></td>\n",
       "      <td>games_yosha/gamedc7dd30a-5d00-45a9-830c-5e485d...</td>\n",
       "      <td></td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>26</th>\n",
       "      <td>Stockfish</td>\n",
       "      <td>gpt-3.5-turbo-instruct</td>\n",
       "      <td>1694</td>\n",
       "      <td>?</td>\n",
       "      <td>*</td>\n",
       "      <td>85</td>\n",
       "      <td>1</td>\n",
       "      <td>[White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>False</td>\n",
       "      <td>True</td>\n",
       "      <td>0-1</td>\n",
       "      <td>games_yosha/gamec797972f-2990-4611-a312-4433dd...</td>\n",
       "      <td></td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>25</th>\n",
       "      <td>Stockfish</td>\n",
       "      <td>gpt-3.5-turbo-instruct</td>\n",
       "      <td>1694</td>\n",
       "      <td>?</td>\n",
       "      <td>0-1</td>\n",
       "      <td>100</td>\n",
       "      <td>1</td>\n",
       "      <td>[White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td></td>\n",
       "      <td>games_yosha/game22edbe69-afb6-417b-98a2-5035be...</td>\n",
       "      <td></td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>24</th>\n",
       "      <td>Stockfish</td>\n",
       "      <td>gpt-3.5-turbo-instruct</td>\n",
       "      <td>1694</td>\n",
       "      <td>?</td>\n",
       "      <td>1-0</td>\n",
       "      <td>77</td>\n",
       "      <td>1</td>\n",
       "      <td>[White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td></td>\n",
       "      <td>games_yosha/gamef268c8e6-a4ec-419a-a131-87e118...</td>\n",
       "      <td></td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>62</th>\n",
       "      <td>Stockfish</td>\n",
       "      <td>gpt-3.5-turbo-instruct</td>\n",
       "      <td>1954</td>\n",
       "      <td>?</td>\n",
       "      <td>1-0</td>\n",
       "      <td>221</td>\n",
       "      <td>1</td>\n",
       "      <td>[White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td></td>\n",
       "      <td>games_yosha/gamec5b08257-48d7-48f3-8f2a-29d5a6...</td>\n",
       "      <td></td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>61</th>\n",
       "      <td>Stockfish</td>\n",
       "      <td>gpt-3.5-turbo-instruct</td>\n",
       "      <td>1954</td>\n",
       "      <td>?</td>\n",
       "      <td>0-1</td>\n",
       "      <td>78</td>\n",
       "      <td>1</td>\n",
       "      <td>[White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td></td>\n",
       "      <td>games_yosha/gamec10b7b05-1362-41b6-a170-3191df...</td>\n",
       "      <td></td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>79</th>\n",
       "      <td>Stockfish</td>\n",
       "      <td>gpt-3.5-turbo-instruct</td>\n",
       "      <td>1954</td>\n",
       "      <td>?</td>\n",
       "      <td>1/2-1/2</td>\n",
       "      <td>161</td>\n",
       "      <td>1</td>\n",
       "      <td>[White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td></td>\n",
       "      <td>games_yosha/gamea541823d-a8bb-45fa-af00-adfafc...</td>\n",
       "      <td></td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>69</th>\n",
       "      <td>Stockfish</td>\n",
       "      <td>gpt-3.5-turbo-instruct</td>\n",
       "      <td>1954</td>\n",
       "      <td>?</td>\n",
       "      <td>1/2-1/2</td>\n",
       "      <td>147</td>\n",
       "      <td>1</td>\n",
       "      <td>[White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td></td>\n",
       "      <td>games_yosha/game7cfbec2c-0e4d-4bf9-8a1f-d85c78...</td>\n",
       "      <td></td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>80</th>\n",
       "      <td>Stockfish</td>\n",
       "      <td>gpt-3.5-turbo-instruct</td>\n",
       "      <td>1954</td>\n",
       "      <td>?</td>\n",
       "      <td>1-0</td>\n",
       "      <td>51</td>\n",
       "      <td>1</td>\n",
       "      <td>[White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td></td>\n",
       "      <td>games_yosha/game251270b2-0667-44e8-b49b-384e16...</td>\n",
       "      <td></td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>81 rows × 15 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "        white                   black white_elo black_elo   result nmoves  \\\n",
       "14  Stockfish  gpt-3.5-turbo-instruct      1694         ?  1/2-1/2    138   \n",
       "27  Stockfish  gpt-3.5-turbo-instruct      1694         ?      1-0     69   \n",
       "26  Stockfish  gpt-3.5-turbo-instruct      1694         ?        *     85   \n",
       "25  Stockfish  gpt-3.5-turbo-instruct      1694         ?      0-1    100   \n",
       "24  Stockfish  gpt-3.5-turbo-instruct      1694         ?      1-0     77   \n",
       "..        ...                     ...       ...       ...      ...    ...   \n",
       "62  Stockfish  gpt-3.5-turbo-instruct      1954         ?      1-0    221   \n",
       "61  Stockfish  gpt-3.5-turbo-instruct      1954         ?      0-1     78   \n",
       "79  Stockfish  gpt-3.5-turbo-instruct      1954         ?  1/2-1/2    161   \n",
       "69  Stockfish  gpt-3.5-turbo-instruct      1954         ?  1/2-1/2    147   \n",
       "80  Stockfish  gpt-3.5-turbo-instruct      1954         ?      1-0     51   \n",
       "\n",
       "   nstarting_move                                           pgn_base  \\\n",
       "14              1  [White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...   \n",
       "27              1  [White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...   \n",
       "26              1  [White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...   \n",
       "25              1  [White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...   \n",
       "24              1  [White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...   \n",
       "..            ...                                                ...   \n",
       "62              1  [White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...   \n",
       "61              1  [White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...   \n",
       "79              1  [White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...   \n",
       "69              1  [White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...   \n",
       "80              1  [White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...   \n",
       "\n",
       "    temperature random_engine has_illegal illegal_move  \\\n",
       "14          0.0         False       False                \n",
       "27          0.0         False       False                \n",
       "26          0.0         False        True          0-1   \n",
       "25          0.0         False       False                \n",
       "24          0.0         False       False                \n",
       "..          ...           ...         ...          ...   \n",
       "62          0.0         False       False                \n",
       "61          0.0         False       False                \n",
       "79          0.0         False       False                \n",
       "69          0.0         False       False                \n",
       "80          0.0         False       False                \n",
       "\n",
       "                                          folder_name pgn_base_moves  \\\n",
       "14  games_yosha/gamea4187bf2-f3e0-44b2-ab5c-91aedf...                  \n",
       "27  games_yosha/gamedc7dd30a-5d00-45a9-830c-5e485d...                  \n",
       "26  games_yosha/gamec797972f-2990-4611-a312-4433dd...                  \n",
       "25  games_yosha/game22edbe69-afb6-417b-98a2-5035be...                  \n",
       "24  games_yosha/gamef268c8e6-a4ec-419a-a131-87e118...                  \n",
       "..                                                ...            ...   \n",
       "62  games_yosha/gamec5b08257-48d7-48f3-8f2a-29d5a6...                  \n",
       "61  games_yosha/gamec10b7b05-1362-41b6-a170-3191df...                  \n",
       "79  games_yosha/gamea541823d-a8bb-45fa-af00-adfafc...                  \n",
       "69  games_yosha/game7cfbec2c-0e4d-4bf9-8a1f-d85c78...                  \n",
       "80  games_yosha/game251270b2-0667-44e8-b49b-384e16...                  \n",
       "\n",
       "    base_pgn_prompt  \n",
       "14            False  \n",
       "27            False  \n",
       "26            False  \n",
       "25            False  \n",
       "24            False  \n",
       "..              ...  \n",
       "62            False  \n",
       "61            False  \n",
       "79            False  \n",
       "69            False  \n",
       "80            False  \n",
       "\n",
       "[81 rows x 15 columns]"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df.sort_values(by=['white_elo'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "bf38750b",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "48693dae",
   "metadata": {},
   "outputs": [],
   "source": [
    "\n",
    "\n",
    "\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "c9eee201",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Analysis against SF (no random engine, no random first moves) for model: gpt-3.5-turbo-instruct\n",
      "Score 40.97 % for games with only legal moves\n",
      "Score 29.5 for games with only legal moves ( 72 games)\n",
      "Score 36.42 % for all games, being legal or illegal\n",
      "Score 29.5 for all games ( 81 games)\n",
      "Out of 81 games against SF, 72 were legal games and 9 were illegal games, hence 11 % of illegal games.\n",
      "72 legal games and 9 illegal games (out of 81 total games)\n",
      "11 % of illegal games\n",
      "Illegal moves are:\n",
      "|     |   illegal_move |\n",
      "|:----|---------------:|\n",
      "| 0-1 |              9 |\n",
      "Illegal moves with n first random move are:\n",
      "| illegal_move   |\n",
      "|----------------|\n"
     ]
    }
   ],
   "source": [
    "df_non_random = df.query(\"random_engine == False and nstarting_move <= 2\")\n",
    "\n",
    "def analyze_model_performance(df_m, model_gpt_name):    \n",
    "\n",
    "    print(f\"Analysis against SF (no random engine, no random first moves) for model: {model_gpt_name}\")\n",
    "\n",
    "    model_games = df_m.query(f\"(white == '{model_gpt_name}' or black == '{model_gpt_name}') and has_illegal == False\")\n",
    "    score_model = compute_score(model_games, model_name=model_gpt_name)\n",
    "    print(\"Score\", round(100*score_model, 2), \"% for games with only legal moves\")\n",
    "    score, t = compute_score(model_games, model_name=model_gpt_name, percentage=False)\n",
    "    print(\"Score\", score, \"for games with only legal moves (\", t, \"games)\")\n",
    "\n",
    "    tot = len(df_m.query(f\"(white == '{model_gpt_name}' or black == '{model_gpt_name}')\"))\n",
    "    print(\"Score\", round(100*(score/tot), 2), \"% for all games, being legal or illegal\")\n",
    "    print(\"Score\", score, \"for all games (\", tot, \"games)\")\n",
    "\n",
    "    vals_model = df_m.query(f\"(white == '{model_gpt_name}' or black == '{model_gpt_name}')\")['has_illegal'].value_counts()\n",
    "    ntot = vals_model.get(0, 0) + vals_model.get(1, 0)\n",
    "\n",
    "    print(\"Out of\", ntot ,\"games against SF,\", vals_model.get(0, 0), \"were legal games and\", vals_model.get(1, 0), \"were illegal games, hence\", round((vals_model.get(1, 0)/ntot)*100), \"% of illegal games.\")\n",
    "    print(vals_model.get(0, 0), \"legal games and\", vals_model.get(1, 0), \"illegal games\", \"(out of\", ntot, \"total games)\")\n",
    "    print(round((vals_model.get(1, 0)/ntot)*100), \"% of illegal games\")\n",
    "\n",
    "    print(\"Illegal moves are:\")\n",
    "    print(df_m.query(f\"(white == '{model_gpt_name}' or black == '{model_gpt_name}') and has_illegal == True\")['illegal_move'].value_counts().to_markdown())\n",
    "\n",
    "    print(\"Illegal moves with n first random move are:\")\n",
    "    print(df_m.query(f\"(white == '{model_gpt_name}' or black == '{model_gpt_name}') and has_illegal == True and nstarting_move > 2\")['illegal_move'].value_counts().to_markdown())\n",
    "\n",
    "# Example usage:\n",
    "analyze_model_performance(df_non_random, 'gpt-3.5-turbo-instruct')\n",
    "# analyze_model_performance(df_non_random, 'gpt-4')\n",
    "\n",
    "# analyze_model_performance(df_random, 'gpt-4')\n",
    "\n",
    "# analyze_model_performance(df_random, 'gpt-3.5-turbo')\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "fb299e27",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "9e5b8c4b",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "00a4a2b7",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "81"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(df_non_random.query(\"(white == 'gpt-3.5-turbo-instruct' or black == 'gpt-3.5-turbo-instruct')\"))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "afa8aafa",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0 illegal moves with white\n",
      "0\n",
      "81\n",
      "9 illegal moves with black\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "(0, 0, 81, 9)"
      ]
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "def count_moves_and_illegals(df_l, gpt_model_name):\n",
    "    illegal_white_moves = len(df_l.query(f\"(white == '{gpt_model_name}') and has_illegal == True\"))\n",
    "    total_white_moves = len(df_l.query(f\"(white == '{gpt_model_name}')\"))\n",
    "    total_black_moves = len(df_l.query(f\"(black == '{gpt_model_name}')\"))\n",
    "    illegal_black_moves = len(df_l.query(f\"(black == '{gpt_model_name}') and has_illegal == True\"))\n",
    "\n",
    "    print(illegal_white_moves, \"illegal moves with white\")\n",
    "    print(total_white_moves)\n",
    "    print(total_black_moves)\n",
    "    print(illegal_black_moves, \"illegal moves with black\")\n",
    "\n",
    "    return illegal_white_moves, total_white_moves, total_black_moves, illegal_black_moves\n",
    "\n",
    "count_moves_and_illegals(df_non_random, 'gpt-3.5-turbo-instruct')\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "d6b3ffaf",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "temperature\n",
      "0.0            9\n",
      "dtype: int64\n",
      "0.0 % of games with temperature 0.8\n",
      "| illegal_move   |\n",
      "|----------------|\n"
     ]
    }
   ],
   "source": [
    "mask = (df_non_random[\"white\"] == 'gpt-3.5-turbo-instruct') | (df_non_random[\"black\"] == 'gpt-3.5-turbo-instruct')\n",
    "illegal_gpt35vsSF_instruct = df_non_random.loc[mask & (df_non_random[\"has_illegal\"] == True)].copy()\n",
    "illegal_gpt35vsSF_instruct[\"temperature\"] = illegal_gpt35vsSF_instruct[\"temperature\"].astype(float)\n",
    "\n",
    "print(illegal_gpt35vsSF_instruct[[\"temperature\"]].value_counts())\n",
    "\n",
    "t8 = len(df_non_random.query(\"(white == 'gpt-3.5-turbo-instruct' or black == 'gpt-3.5-turbo-instruct') and `temperature` == '0.8'\")) / len(df_non_random.query(\"(white == 'gpt-3.5-turbo-instruct' or black == 'gpt-3.5-turbo-instruct')\"))\n",
    "print(round(100*t8, 2), \"% of games with temperature 0.8\")\n",
    "\n",
    "print(df_non_random.query(\"(white == 'gpt-3.5-turbo-instruct' or black == 'gpt-3.5-turbo-instruct') and `temperature` == '0.8' and has_illegal == True\")['illegal_move'].value_counts().to_markdown())\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "bb8fe079",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0 games with altered prompt\n",
      "0.0 % games with altered prompt\n",
      "0 illegal games with altered prompts and 9 illegal games with original prompts\n",
      "Series([], Name: illegal_move, dtype: int64)\n",
      "|    |   nstarting_move |\n",
      "|---:|-----------------:|\n",
      "|  1 |                9 |\n"
     ]
    }
   ],
   "source": [
    "illegal_gpt35vsSF_instruct = df_non_random.query(\"(white == 'gpt-3.5-turbo-instruct' or black == 'gpt-3.5-turbo-instruct') and has_illegal == True\") # [['pgn_base']].value_counts()\n",
    "n = 0\n",
    "m = 0\n",
    "for i, row in illegal_gpt35vsSF_instruct.iterrows():\n",
    "    if \"Rennes FRA\" in row[\"pgn_base\"]:\n",
    "        n = n + 1\n",
    "    else:\n",
    "        m = m + 1\n",
    "\n",
    "print(len(df_non_random.query(\"(white == 'gpt-3.5-turbo-instruct' or black == 'gpt-3.5-turbo-instruct') and pgn_base.str.contains('Rennes FRA')\")), \"games with altered prompt\")\n",
    "print(len(df_non_random.query(\"(white == 'gpt-3.5-turbo-instruct' or black == 'gpt-3.5-turbo-instruct') and pgn_base.str.contains('Rennes FRA')\")) / len(df_non_random.query(\"(white == 'gpt-3.5-turbo-instruct' or black == 'gpt-3.5-turbo-instruct')\")) * 100, \"% games with altered prompt\")\n",
    "\n",
    "print(n, \"illegal games with altered prompts and\", m, \"illegal games with original prompts\")\n",
    "print(illegal_gpt35vsSF_instruct.query(\"pgn_base.str.contains('Rennes FRA')\")['illegal_move'].value_counts())\n",
    "\n",
    "print(df_non_random.query(\"(white == 'gpt-3.5-turbo-instruct' or black == 'gpt-3.5-turbo-instruct') and has_illegal == True\")['nstarting_move'].value_counts().to_markdown())\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "9b7ee51b",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Does random first moves impact?\n",
      "|    |   nstarting_move |\n",
      "|---:|-----------------:|\n",
      "|  1 |               81 |\n"
     ]
    }
   ],
   "source": [
    "print(\"Does random first moves impact?\")\n",
    "print(df_non_random.query(\"(white == 'gpt-3.5-turbo-instruct' or black == 'gpt-3.5-turbo-instruct')\")['nstarting_move'].value_counts().to_markdown())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "id": "93c63170",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0-1 games_yosha/gameac4eefd4-b0a5-4289-84e9-7b2f93223f54\n",
      "0-1 games_yosha/game31767cf0-7145-4798-af1a-cbc99d65ace8\n",
      "0-1 games_yosha/gamec797972f-2990-4611-a312-4433ddd56fb5\n",
      "0-1 games_yosha/gamee099abf1-e680-4aa1-a429-3247c1b772cf\n",
      "0-1 games_yosha/game45307c19-414f-4671-a6be-9f4e624af391\n",
      "0-1 games_yosha/gamebf09a82a-7b2d-4816-b125-afb87b6eaeb1\n",
      "0-1 games_yosha/gamefa4d9bcd-9932-49bf-8880-c439fd7ea1fd\n",
      "0-1 games_yosha/gamea0d12b34-eb64-44b5-83e9-26ae152fbd6e\n",
      "0-1 games_yosha/game6372e7a4-c214-4386-85e5-2ee5f38d32c9\n",
      "If we consider \"1-0\", \"comments\", and \"unspecified promotions\" as fixable, then 11.1 % are non valid games\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "11.1"
      ]
     },
     "execution_count": 19,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "def count_invalid_games(df_l, model_gpt_name):\n",
    "    query_str = f\"(white == '{model_gpt_name}' or black == '{model_gpt_name}') and has_illegal == True and illegal_move != '1-0' and illegal_move != '1-'\"\n",
    "    \n",
    "    x = 0\n",
    "    for _, r in df_l.query(query_str).iterrows():\n",
    "        if '=' in r['illegal_move'] or '{' in r['illegal_move']:\n",
    "            continue\n",
    "\n",
    "        print(r['illegal_move'], r['folder_name'])\n",
    "        x += 1 \n",
    "\n",
    "    percentage = round(x / len(df_l.query(f\"(white == '{model_gpt_name}' or black == '{model_gpt_name}')\")), 3) * 100\n",
    "    print(f'If we consider \"1-0\", \"comments\", and \"unspecified promotions\" as fixable, then {percentage} % are non valid games')\n",
    "\n",
    "    return percentage\n",
    "\n",
    "count_invalid_games(df_non_random, 'gpt-3.5-turbo-instruct')\n",
    "\n",
    "\n",
    "    "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "id": "f2b5d116",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Qualitative analysis of illegal moves:\n",
      "\n"
     ]
    }
   ],
   "source": [
    "print(\"Qualitative analysis of illegal moves:\")\n",
    "\n",
    "import pandas as pd\n",
    "from stockfish import Stockfish\n",
    "\n",
    "# A list to store your data\n",
    "data_list = []\n",
    "\n",
    "illegal_moves10 = df_non_random.query(\"(white == 'gpt-3.5-turbo-instruct' or black == 'gpt-3.5-turbo-instruct') and illegal_move == '1-0'\")\n",
    "\n",
    "for index, row in illegal_moves10.iterrows():\n",
    "    if row['white'] == 'gpt-3.5-turbo-instruct':\n",
    "        color = \"White\"\n",
    "    else:\n",
    "        color = \"Black\"\n",
    "\n",
    "    # read the PGN file\n",
    "    with open(row['folder_name'] + \"/game.pgn\") as pgn:\n",
    "        game = chess.pgn.read_game(pgn)\n",
    "        board = game.board()\n",
    "        for move in game.mainline_moves():\n",
    "            board.push(move)\n",
    "\n",
    "        stockfish = Stockfish(\"./stockfish/stockfish/stockfish-ubuntu-x86-64-avx2\")\n",
    "        stockfish.set_position([str(m) for m in game.mainline_moves()])\n",
    "        # stockfish._go_time(5000)\n",
    "        ev = stockfish.get_evaluation()\n",
    "\n",
    "        # Create an evaluation string\n",
    "        if ev['type'] == 'cp':\n",
    "            evaluation = str(ev['value']/100)\n",
    "        else:\n",
    "            evaluation = \"Mate in \" + str(ev['value'])\n",
    "\n",
    "        # Append the dictionary to your list\n",
    "        data_list.append({\n",
    "            \"GPT Color\": color,\n",
    "            \"Assessment\": evaluation\n",
    "        })\n",
    "\n",
    "# Convert the list of dictionaries into a dataframe\n",
    "df_results = pd.DataFrame(data_list)\n",
    "\n",
    "# If you want to see the first few rows of your dataframe:\n",
    "print(df_results.to_markdown())\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "id": "a6d7509c",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>white</th>\n",
       "      <th>black</th>\n",
       "      <th>white_elo</th>\n",
       "      <th>black_elo</th>\n",
       "      <th>result</th>\n",
       "      <th>nmoves</th>\n",
       "      <th>nstarting_move</th>\n",
       "      <th>pgn_base</th>\n",
       "      <th>temperature</th>\n",
       "      <th>random_engine</th>\n",
       "      <th>has_illegal</th>\n",
       "      <th>illegal_move</th>\n",
       "      <th>folder_name</th>\n",
       "      <th>pgn_base_moves</th>\n",
       "      <th>base_pgn_prompt</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>10</th>\n",
       "      <td>Stockfish</td>\n",
       "      <td>gpt-3.5-turbo-instruct</td>\n",
       "      <td>1785</td>\n",
       "      <td>?</td>\n",
       "      <td>*</td>\n",
       "      <td>139</td>\n",
       "      <td>1</td>\n",
       "      <td>[White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>False</td>\n",
       "      <td>True</td>\n",
       "      <td>0-1</td>\n",
       "      <td>games_yosha/game31767cf0-7145-4798-af1a-cbc99d...</td>\n",
       "      <td></td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>72</th>\n",
       "      <td>Stockfish</td>\n",
       "      <td>gpt-3.5-turbo-instruct</td>\n",
       "      <td>1954</td>\n",
       "      <td>?</td>\n",
       "      <td>*</td>\n",
       "      <td>113</td>\n",
       "      <td>1</td>\n",
       "      <td>[White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>False</td>\n",
       "      <td>True</td>\n",
       "      <td>0-1</td>\n",
       "      <td>games_yosha/gamea0d12b34-eb64-44b5-83e9-26ae15...</td>\n",
       "      <td></td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>43</th>\n",
       "      <td>Stockfish</td>\n",
       "      <td>gpt-3.5-turbo-instruct</td>\n",
       "      <td>1871</td>\n",
       "      <td>?</td>\n",
       "      <td>*</td>\n",
       "      <td>111</td>\n",
       "      <td>1</td>\n",
       "      <td>[White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>False</td>\n",
       "      <td>True</td>\n",
       "      <td>0-1</td>\n",
       "      <td>games_yosha/game45307c19-414f-4671-a6be-9f4e62...</td>\n",
       "      <td></td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>46</th>\n",
       "      <td>Stockfish</td>\n",
       "      <td>gpt-3.5-turbo-instruct</td>\n",
       "      <td>1871</td>\n",
       "      <td>?</td>\n",
       "      <td>*</td>\n",
       "      <td>109</td>\n",
       "      <td>1</td>\n",
       "      <td>[White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>False</td>\n",
       "      <td>True</td>\n",
       "      <td>0-1</td>\n",
       "      <td>games_yosha/gamefa4d9bcd-9932-49bf-8880-c439fd...</td>\n",
       "      <td></td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>30</th>\n",
       "      <td>Stockfish</td>\n",
       "      <td>gpt-3.5-turbo-instruct</td>\n",
       "      <td>1694</td>\n",
       "      <td>?</td>\n",
       "      <td>*</td>\n",
       "      <td>99</td>\n",
       "      <td>1</td>\n",
       "      <td>[White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>False</td>\n",
       "      <td>True</td>\n",
       "      <td>0-1</td>\n",
       "      <td>games_yosha/gamee099abf1-e680-4aa1-a429-3247c1...</td>\n",
       "      <td></td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>Stockfish</td>\n",
       "      <td>gpt-3.5-turbo-instruct</td>\n",
       "      <td>1785</td>\n",
       "      <td>?</td>\n",
       "      <td>*</td>\n",
       "      <td>91</td>\n",
       "      <td>1</td>\n",
       "      <td>[White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>False</td>\n",
       "      <td>True</td>\n",
       "      <td>0-1</td>\n",
       "      <td>games_yosha/gameac4eefd4-b0a5-4289-84e9-7b2f93...</td>\n",
       "      <td></td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>26</th>\n",
       "      <td>Stockfish</td>\n",
       "      <td>gpt-3.5-turbo-instruct</td>\n",
       "      <td>1694</td>\n",
       "      <td>?</td>\n",
       "      <td>*</td>\n",
       "      <td>85</td>\n",
       "      <td>1</td>\n",
       "      <td>[White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>False</td>\n",
       "      <td>True</td>\n",
       "      <td>0-1</td>\n",
       "      <td>games_yosha/gamec797972f-2990-4611-a312-4433dd...</td>\n",
       "      <td></td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>44</th>\n",
       "      <td>Stockfish</td>\n",
       "      <td>gpt-3.5-turbo-instruct</td>\n",
       "      <td>1871</td>\n",
       "      <td>?</td>\n",
       "      <td>*</td>\n",
       "      <td>85</td>\n",
       "      <td>1</td>\n",
       "      <td>[White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>False</td>\n",
       "      <td>True</td>\n",
       "      <td>0-1</td>\n",
       "      <td>games_yosha/gamebf09a82a-7b2d-4816-b125-afb87b...</td>\n",
       "      <td></td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>75</th>\n",
       "      <td>Stockfish</td>\n",
       "      <td>gpt-3.5-turbo-instruct</td>\n",
       "      <td>1954</td>\n",
       "      <td>?</td>\n",
       "      <td>*</td>\n",
       "      <td>65</td>\n",
       "      <td>1</td>\n",
       "      <td>[White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>False</td>\n",
       "      <td>True</td>\n",
       "      <td>0-1</td>\n",
       "      <td>games_yosha/game6372e7a4-c214-4386-85e5-2ee5f3...</td>\n",
       "      <td></td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "        white                   black white_elo black_elo result nmoves  \\\n",
       "10  Stockfish  gpt-3.5-turbo-instruct      1785         ?      *    139   \n",
       "72  Stockfish  gpt-3.5-turbo-instruct      1954         ?      *    113   \n",
       "43  Stockfish  gpt-3.5-turbo-instruct      1871         ?      *    111   \n",
       "46  Stockfish  gpt-3.5-turbo-instruct      1871         ?      *    109   \n",
       "30  Stockfish  gpt-3.5-turbo-instruct      1694         ?      *     99   \n",
       "4   Stockfish  gpt-3.5-turbo-instruct      1785         ?      *     91   \n",
       "26  Stockfish  gpt-3.5-turbo-instruct      1694         ?      *     85   \n",
       "44  Stockfish  gpt-3.5-turbo-instruct      1871         ?      *     85   \n",
       "75  Stockfish  gpt-3.5-turbo-instruct      1954         ?      *     65   \n",
       "\n",
       "   nstarting_move                                           pgn_base  \\\n",
       "10              1  [White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...   \n",
       "72              1  [White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...   \n",
       "43              1  [White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...   \n",
       "46              1  [White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...   \n",
       "30              1  [White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...   \n",
       "4               1  [White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...   \n",
       "26              1  [White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...   \n",
       "44              1  [White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...   \n",
       "75              1  [White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...   \n",
       "\n",
       "    temperature random_engine has_illegal illegal_move  \\\n",
       "10          0.0         False        True          0-1   \n",
       "72          0.0         False        True          0-1   \n",
       "43          0.0         False        True          0-1   \n",
       "46          0.0         False        True          0-1   \n",
       "30          0.0         False        True          0-1   \n",
       "4           0.0         False        True          0-1   \n",
       "26          0.0         False        True          0-1   \n",
       "44          0.0         False        True          0-1   \n",
       "75          0.0         False        True          0-1   \n",
       "\n",
       "                                          folder_name pgn_base_moves  \\\n",
       "10  games_yosha/game31767cf0-7145-4798-af1a-cbc99d...                  \n",
       "72  games_yosha/gamea0d12b34-eb64-44b5-83e9-26ae15...                  \n",
       "43  games_yosha/game45307c19-414f-4671-a6be-9f4e62...                  \n",
       "46  games_yosha/gamefa4d9bcd-9932-49bf-8880-c439fd...                  \n",
       "30  games_yosha/gamee099abf1-e680-4aa1-a429-3247c1...                  \n",
       "4   games_yosha/gameac4eefd4-b0a5-4289-84e9-7b2f93...                  \n",
       "26  games_yosha/gamec797972f-2990-4611-a312-4433dd...                  \n",
       "44  games_yosha/gamebf09a82a-7b2d-4816-b125-afb87b...                  \n",
       "75  games_yosha/game6372e7a4-c214-4386-85e5-2ee5f3...                  \n",
       "\n",
       "    base_pgn_prompt  \n",
       "10            False  \n",
       "72            False  \n",
       "43            False  \n",
       "46            False  \n",
       "30            False  \n",
       "4             False  \n",
       "26            False  \n",
       "44            False  \n",
       "75            False  "
      ]
     },
     "execution_count": 21,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_non_random.query(\"(white == 'gpt-3.5-turbo-instruct' or black == 'gpt-3.5-turbo-instruct' and has_illegal == True)\").sort_values(by=\"nmoves\", ascending=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "id": "bb15fb4a",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0 games with mate out of 0 games\n",
      "0 games with cp > 5.0 out of 0 games\n",
      "0 games with cp < 5.0 out of 0 games\n"
     ]
    }
   ],
   "source": [
    "j = 0\n",
    "cp_g = 0\n",
    "for index, row in df_results.iterrows():\n",
    "    if 'Mate' in row['Assessment']:\n",
    "        j = j + 1\n",
    "    elif float(row['Assessment']) > 5.0:\n",
    "        cp_g = cp_g + 1\n",
    "\n",
    "print(j, \"games with mate\", \"out of\", len(df_results), \"games\")\n",
    "print(cp_g, \"games with cp > 5.0\", \"out of\", len(df_results), \"games\")\n",
    "print(len(df_results) - (j+cp_g), \"games with cp < 5.0\", \"out of\", len(df_results), \"games\")\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "id": "c3d4c710",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "illegal moves 0.21 %\n",
      "illegal_moves without 1-0: 0.21 %\n",
      "Total number of moves played by gpt-3.5-turbo-instruct : 4282.0\n",
      "White played 0.0 moves\n",
      "Black played 4282.0 moves\n",
      "Number of moves against SF\n",
      "The longest game was 126.5 moves\n",
      "The shortest game was 14.0 moves\n",
      "The average game length was 52.864197530864196 moves\n",
      "The median game length was 49.5 moves\n"
     ]
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 720x432 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "\n",
    "def analyze_games(df_l, model_name):\n",
    "    # Query for games played by the given model\n",
    "    model_games = df_l.query(f\"(white == '{model_name}' or black == '{model_name}')\")\n",
    "    model_games['nmoves'] = model_games['nmoves'] / 2\n",
    "    \n",
    "    # Sum of moves played by the model\n",
    "    white_nmoves = df_l.query(f\"white == '{model_name}'\")['nmoves'].sum() / 2\n",
    "    black_nmoves = df_l.query(f\"black == '{model_name}'\")['nmoves'].sum() / 2\n",
    "    t_moves = model_games['nmoves'].sum()\n",
    "\n",
    "    # Count of illegal moves\n",
    "    illegal_moves = len(df_l.query(f\"(white == '{model_name}' or black == '{model_name}') and has_illegal == True\"))\n",
    "    illegal_moves_10 = len(df_l.query(f\"(white == '{model_name}' or black == '{model_name}') and has_illegal == True and illegal_move != '1-0'\"))\n",
    "\n",
    "    # Printing results\n",
    "    print(\"illegal moves\", round(illegal_moves/t_moves, 5)*100, \"%\")\n",
    "    print(\"illegal_moves without 1-0:\", round(illegal_moves_10/t_moves, 5)*100, \"%\")\n",
    "    print(\"Total number of moves played by\", model_name, \":\", t_moves)\n",
    "    print(\"White played\", white_nmoves, \"moves\")\n",
    "    print(\"Black played\", black_nmoves, \"moves\")\n",
    "    print(\"Number of moves against SF\")\n",
    "    print(\"The longest game was\", model_games['nmoves'].max(), \"moves\")\n",
    "    print(\"The shortest game was\", model_games['nmoves'].min(), \"moves\")\n",
    "    print(\"The average game length was\", model_games['nmoves'].mean(), \"moves\")\n",
    "    print(\"The median game length was\", model_games['nmoves'].median(), \"moves\")\n",
    "\n",
    "    # Plotting box plot\n",
    "    plt.figure(figsize=(10, 6))\n",
    "    plt.boxplot(model_games['nmoves'])\n",
    "    plt.title(f\"Number of moves per game against SF\")\n",
    "    plt.ylabel(\"Number of moves\")\n",
    "    plt.xlabel(f\"{model_name}\")\n",
    "    plt.savefig(f\"{model_name}_games_nmoves.png\", dpi=300, bbox_inches='tight')\n",
    "    plt.show()\n",
    "\n",
    "# Example usage:\n",
    "analyze_games(df_non_random, 'gpt-3.5-turbo-instruct')\n",
    "# analyze_games(df_non_random, 'gpt-4')\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "3acae5ba",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "id": "57be0777",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "81 games played against SF\n",
      "72 legal games played against SF\n"
     ]
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 720x360 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "df_gpt35 = df_non_random.query(\"(white == 'gpt-3.5-turbo-instruct' or black == 'gpt-3.5-turbo-instruct')\")\n",
    "print(len(df_gpt35), \"games played against SF\")\n",
    "print(len(df_gpt35.query(\"has_illegal == False\")), \"legal games played against SF\")\n",
    "\n",
    "# Extracting elo values for games where Stockfish is playing against gpt-3.5-turbo-instruct\n",
    "stockfish_black_elo_vs_gpt3_5_instruct = df_gpt35['black_elo']\n",
    "# remove values with '?' value\n",
    "stockfish_black_elo_vs_gpt3_5_instruct = stockfish_black_elo_vs_gpt3_5_instruct[stockfish_black_elo_vs_gpt3_5_instruct != '?']\n",
    "\n",
    "stockfish_white_elo_vs_gpt3_5_instruct = df_gpt35['white_elo']\n",
    "# remove values with '?' value\n",
    "stockfish_white_elo_vs_gpt3_5_instruct = stockfish_white_elo_vs_gpt3_5_instruct[stockfish_white_elo_vs_gpt3_5_instruct != '?']\n",
    "\n",
    "\n",
    "# Combining Elo ratings of Stockfish as both white and black player\n",
    "combined_stockfish_elo = pd.concat([stockfish_white_elo_vs_gpt3_5_instruct, stockfish_black_elo_vs_gpt3_5_instruct])\n",
    "\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "# Assuming you have already computed value counts\n",
    "value_counts = combined_stockfish_elo.value_counts().sort_index()\n",
    "\n",
    "# Plotting the histogram for combined Elo ratings\n",
    "plt.figure(figsize=(10,5))\n",
    "plt.bar(value_counts.index, value_counts.values)\n",
    "plt.xlabel('Elo rating')\n",
    "plt.ylabel('Number of games')\n",
    "plt.title('Elo rating distribution of Stockfish for games of gpt-3.5-turbo-instruct against Stockfish')\n",
    "plt.savefig(\"stockfish_elo_distribution.png\", dpi=300, bbox_inches='tight')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "b621c14c",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "id": "93722364",
   "metadata": {},
   "outputs": [],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "import seaborn as sns\n",
    "\n",
    "# Set the style of seaborn for better visualization\n",
    "sns.set(style=\"whitegrid\")\n",
    "\n",
    "def plot_histogram(df):\n",
    "    # Initialize the matplotlib figure\n",
    "    plt.figure(figsize=(12, 6))\n",
    "\n",
    "    # Plot histogram of nmoves\n",
    "    sns.histplot(df['nmoves'], bins=30, kde=True, color=\"skyblue\")\n",
    "\n",
    "    # Set plot labels and title\n",
    "    plt.xlabel('Number of Moves', fontsize=14)\n",
    "    plt.ylabel('Frequency', fontsize=14)\n",
    "    plt.title('Distribution of Number of Moves in Chess Games', fontsize=16)\n",
    "\n",
    "    # Show the plot\n",
    "    plt.show()\n",
    "\n",
    "def plot_boxplot(df):\n",
    "    # Initialize the matplotlib figure\n",
    "    plt.figure(figsize=(10, 6))\n",
    "\n",
    "    # Create a boxplot of nmoves\n",
    "    sns.boxplot(x=df['nmoves'], color=\"lightblue\")\n",
    "\n",
    "    # Set plot labels and title\n",
    "    plt.xlabel('Number of Moves', fontsize=14)\n",
    "    plt.title('Boxplot of Number of Moves in Chess Games', fontsize=16)\n",
    "\n",
    "    # Show the plot\n",
    "    plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "7a1f1abd",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "d1701256",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "7549abf5",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "8ace2bf9",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "a575c38f",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "8715692d",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "id": "e7445721",
   "metadata": {},
   "outputs": [],
   "source": [
    "df = df.query(\"random_engine == False\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "d40acc82",
   "metadata": {},
   "outputs": [],
   "source": [
    "\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "id": "de5a2b46",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([0.])"
      ]
     },
     "execution_count": 27,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df['temperature'].unique()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "id": "dc0a104c",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>white</th>\n",
       "      <th>black</th>\n",
       "      <th>white_elo</th>\n",
       "      <th>black_elo</th>\n",
       "      <th>result</th>\n",
       "      <th>nmoves</th>\n",
       "      <th>nstarting_move</th>\n",
       "      <th>pgn_base</th>\n",
       "      <th>temperature</th>\n",
       "      <th>random_engine</th>\n",
       "      <th>has_illegal</th>\n",
       "      <th>illegal_move</th>\n",
       "      <th>folder_name</th>\n",
       "      <th>pgn_base_moves</th>\n",
       "      <th>base_pgn_prompt</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>Stockfish</td>\n",
       "      <td>gpt-3.5-turbo-instruct</td>\n",
       "      <td>1785</td>\n",
       "      <td>?</td>\n",
       "      <td>1-0</td>\n",
       "      <td>141</td>\n",
       "      <td>1</td>\n",
       "      <td>[White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td></td>\n",
       "      <td>games_yosha/gameac41f898-b3ca-4b96-90fb-19246e...</td>\n",
       "      <td></td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>Stockfish</td>\n",
       "      <td>gpt-3.5-turbo-instruct</td>\n",
       "      <td>1785</td>\n",
       "      <td>?</td>\n",
       "      <td>0-1</td>\n",
       "      <td>58</td>\n",
       "      <td>1</td>\n",
       "      <td>[White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td></td>\n",
       "      <td>games_yosha/gamecc87b77a-986f-4846-9830-095662...</td>\n",
       "      <td></td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>Stockfish</td>\n",
       "      <td>gpt-3.5-turbo-instruct</td>\n",
       "      <td>1785</td>\n",
       "      <td>?</td>\n",
       "      <td>1-0</td>\n",
       "      <td>53</td>\n",
       "      <td>1</td>\n",
       "      <td>[White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td></td>\n",
       "      <td>games_yosha/gameb0a3fbe2-63a6-4fd4-8a69-605854...</td>\n",
       "      <td></td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>Stockfish</td>\n",
       "      <td>gpt-3.5-turbo-instruct</td>\n",
       "      <td>1785</td>\n",
       "      <td>?</td>\n",
       "      <td>1-0</td>\n",
       "      <td>75</td>\n",
       "      <td>1</td>\n",
       "      <td>[White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td></td>\n",
       "      <td>games_yosha/gamedc31f43d-e0e8-4301-b73f-aed98e...</td>\n",
       "      <td></td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>Stockfish</td>\n",
       "      <td>gpt-3.5-turbo-instruct</td>\n",
       "      <td>1785</td>\n",
       "      <td>?</td>\n",
       "      <td>*</td>\n",
       "      <td>91</td>\n",
       "      <td>1</td>\n",
       "      <td>[White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>False</td>\n",
       "      <td>True</td>\n",
       "      <td>0-1</td>\n",
       "      <td>games_yosha/gameac4eefd4-b0a5-4289-84e9-7b2f93...</td>\n",
       "      <td></td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>76</th>\n",
       "      <td>Stockfish</td>\n",
       "      <td>gpt-3.5-turbo-instruct</td>\n",
       "      <td>1954</td>\n",
       "      <td>?</td>\n",
       "      <td>1-0</td>\n",
       "      <td>115</td>\n",
       "      <td>1</td>\n",
       "      <td>[White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td></td>\n",
       "      <td>games_yosha/gamedc232246-8d2e-42b4-bef2-0cc18a...</td>\n",
       "      <td></td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>77</th>\n",
       "      <td>Stockfish</td>\n",
       "      <td>gpt-3.5-turbo-instruct</td>\n",
       "      <td>1954</td>\n",
       "      <td>?</td>\n",
       "      <td>0-1</td>\n",
       "      <td>88</td>\n",
       "      <td>1</td>\n",
       "      <td>[White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td></td>\n",
       "      <td>games_yosha/gamea8b0b6a6-ad94-45bd-8170-522304...</td>\n",
       "      <td></td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>78</th>\n",
       "      <td>Stockfish</td>\n",
       "      <td>gpt-3.5-turbo-instruct</td>\n",
       "      <td>1954</td>\n",
       "      <td>?</td>\n",
       "      <td>0-1</td>\n",
       "      <td>134</td>\n",
       "      <td>1</td>\n",
       "      <td>[White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td></td>\n",
       "      <td>games_yosha/game8fdf34d5-b73e-4812-8789-d11428...</td>\n",
       "      <td></td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>79</th>\n",
       "      <td>Stockfish</td>\n",
       "      <td>gpt-3.5-turbo-instruct</td>\n",
       "      <td>1954</td>\n",
       "      <td>?</td>\n",
       "      <td>1/2-1/2</td>\n",
       "      <td>161</td>\n",
       "      <td>1</td>\n",
       "      <td>[White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td></td>\n",
       "      <td>games_yosha/gamea541823d-a8bb-45fa-af00-adfafc...</td>\n",
       "      <td></td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>80</th>\n",
       "      <td>Stockfish</td>\n",
       "      <td>gpt-3.5-turbo-instruct</td>\n",
       "      <td>1954</td>\n",
       "      <td>?</td>\n",
       "      <td>1-0</td>\n",
       "      <td>51</td>\n",
       "      <td>1</td>\n",
       "      <td>[White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>False</td>\n",
       "      <td>False</td>\n",
       "      <td></td>\n",
       "      <td>games_yosha/game251270b2-0667-44e8-b49b-384e16...</td>\n",
       "      <td></td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>81 rows × 15 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "        white                   black white_elo black_elo   result nmoves  \\\n",
       "0   Stockfish  gpt-3.5-turbo-instruct      1785         ?      1-0    141   \n",
       "1   Stockfish  gpt-3.5-turbo-instruct      1785         ?      0-1     58   \n",
       "2   Stockfish  gpt-3.5-turbo-instruct      1785         ?      1-0     53   \n",
       "3   Stockfish  gpt-3.5-turbo-instruct      1785         ?      1-0     75   \n",
       "4   Stockfish  gpt-3.5-turbo-instruct      1785         ?        *     91   \n",
       "..        ...                     ...       ...       ...      ...    ...   \n",
       "76  Stockfish  gpt-3.5-turbo-instruct      1954         ?      1-0    115   \n",
       "77  Stockfish  gpt-3.5-turbo-instruct      1954         ?      0-1     88   \n",
       "78  Stockfish  gpt-3.5-turbo-instruct      1954         ?      0-1    134   \n",
       "79  Stockfish  gpt-3.5-turbo-instruct      1954         ?  1/2-1/2    161   \n",
       "80  Stockfish  gpt-3.5-turbo-instruct      1954         ?      1-0     51   \n",
       "\n",
       "   nstarting_move                                           pgn_base  \\\n",
       "0               1  [White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...   \n",
       "1               1  [White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...   \n",
       "2               1  [White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...   \n",
       "3               1  [White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...   \n",
       "4               1  [White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...   \n",
       "..            ...                                                ...   \n",
       "76              1  [White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...   \n",
       "77              1  [White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...   \n",
       "78              1  [White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...   \n",
       "79              1  [White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...   \n",
       "80              1  [White \"Carlsen, Magnus\"]  \\n[Black \"Kasparov,...   \n",
       "\n",
       "    temperature random_engine has_illegal illegal_move  \\\n",
       "0           0.0         False       False                \n",
       "1           0.0         False       False                \n",
       "2           0.0         False       False                \n",
       "3           0.0         False       False                \n",
       "4           0.0         False        True          0-1   \n",
       "..          ...           ...         ...          ...   \n",
       "76          0.0         False       False                \n",
       "77          0.0         False       False                \n",
       "78          0.0         False       False                \n",
       "79          0.0         False       False                \n",
       "80          0.0         False       False                \n",
       "\n",
       "                                          folder_name pgn_base_moves  \\\n",
       "0   games_yosha/gameac41f898-b3ca-4b96-90fb-19246e...                  \n",
       "1   games_yosha/gamecc87b77a-986f-4846-9830-095662...                  \n",
       "2   games_yosha/gameb0a3fbe2-63a6-4fd4-8a69-605854...                  \n",
       "3   games_yosha/gamedc31f43d-e0e8-4301-b73f-aed98e...                  \n",
       "4   games_yosha/gameac4eefd4-b0a5-4289-84e9-7b2f93...                  \n",
       "..                                                ...            ...   \n",
       "76  games_yosha/gamedc232246-8d2e-42b4-bef2-0cc18a...                  \n",
       "77  games_yosha/gamea8b0b6a6-ad94-45bd-8170-522304...                  \n",
       "78  games_yosha/game8fdf34d5-b73e-4812-8789-d11428...                  \n",
       "79  games_yosha/gamea541823d-a8bb-45fa-af00-adfafc...                  \n",
       "80  games_yosha/game251270b2-0667-44e8-b49b-384e16...                  \n",
       "\n",
       "    base_pgn_prompt  \n",
       "0             False  \n",
       "1             False  \n",
       "2             False  \n",
       "3             False  \n",
       "4             False  \n",
       "..              ...  \n",
       "76            False  \n",
       "77            False  \n",
       "78            False  \n",
       "79            False  \n",
       "80            False  \n",
       "\n",
       "[81 rows x 15 columns]"
      ]
     },
     "execution_count": 28,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df.query('nstarting_move <= 2 and temperature == 0.0') # right now nmoves with 1 corresponds to traditional stuff... with 2 corresponds to basic, well-known, random openings and is fair... n > 2 corresponds to k random moves (and is a bit unfair for either side)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "3a044c8e",
   "metadata": {},
   "outputs": [],
   "source": [
    "\n",
    "\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "id": "804c2f9d",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>white</th>\n",
       "      <th>black</th>\n",
       "      <th>white_elo</th>\n",
       "      <th>black_elo</th>\n",
       "      <th>result</th>\n",
       "      <th>nmoves</th>\n",
       "      <th>nstarting_move</th>\n",
       "      <th>pgn_base</th>\n",
       "      <th>temperature</th>\n",
       "      <th>random_engine</th>\n",
       "      <th>has_illegal</th>\n",
       "      <th>illegal_move</th>\n",
       "      <th>folder_name</th>\n",
       "      <th>pgn_base_moves</th>\n",
       "      <th>base_pgn_prompt</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "Empty DataFrame\n",
       "Columns: [white, black, white_elo, black_elo, result, nmoves, nstarting_move, pgn_base, temperature, random_engine, has_illegal, illegal_move, folder_name, pgn_base_moves, base_pgn_prompt]\n",
       "Index: []"
      ]
     },
     "execution_count": 29,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df.query('nstarting_move <= 2 and temperature == 0.8')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "3073946e",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "id": "1d9da2c3",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 864x432 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "\n",
    "\n",
    "plot_histogram(df)\n",
    "# plot_histogram(df.query('nstarting_move > 2'))\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "id": "f8c60a6e",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 720x432 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "\n",
    "\n",
    "plot_boxplot(df)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "id": "a1215102",
   "metadata": {},
   "outputs": [],
   "source": [
    "def compute_average(df, model_name='gpt-3.5-turbo-instruct'):\n",
    "    # Create a copy of the DataFrame to avoid modifying the original\n",
    "    df_copy = df.copy()\n",
    "    \n",
    "    # Convert Elo columns to numeric, setting errors='coerce' to handle non-numeric values\n",
    "    df_copy['white_elo'] = pd.to_numeric(df_copy['white_elo'], errors='coerce')\n",
    "    df_copy['black_elo'] = pd.to_numeric(df_copy['black_elo'], errors='coerce')\n",
    "\n",
    "    # Filter out rows where gpt-3.5-turbo-instruct is the player and get the opponent's Elo\n",
    "    opponent_elo_white = df_copy[df_copy['white'] == model_name]['black_elo']\n",
    "    opponent_elo_black = df_copy[df_copy['black'] == model_name]['white_elo']\n",
    "\n",
    "    # Concatenate the Elo ratings of opponents when gpt-3.5-turbo-instruct played as white and black\n",
    "    all_opponent_elo = pd.concat([opponent_elo_white, opponent_elo_black])\n",
    "\n",
    "    # Calculate the average Elo rating of the opponents, excluding missing or NaN values\n",
    "    average_opponent_elo = all_opponent_elo.mean()\n",
    "\n",
    "    return average_opponent_elo\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "id": "f4ff0c1e",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/mathieuacher/.local/lib/python3.8/site-packages/scipy/optimize/_minpack_py.py:178: RuntimeWarning: The iteration is not making good progress, as measured by the \n",
      "  improvement from the last ten iterations.\n",
      "  warnings.warn(msg, RuntimeWarning)\n",
      "/home/mathieuacher/.local/lib/python3.8/site-packages/scipy/optimize/_minpack_py.py:178: RuntimeWarning: The iteration is not making good progress, as measured by the \n",
      "  improvement from the last ten iterations.\n",
      "  warnings.warn(msg, RuntimeWarning)\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "(1762.0670615316344, 1762.0670615316344, 1700.0, 1700.0)"
      ]
     },
     "execution_count": 33,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from scipy.optimize import fsolve\n",
    "import numpy as np\n",
    "\n",
    "\n",
    "def computation_Elo(df,  initial_guess = 1700, model_name='gpt-3.5-turbo-instruct', K=400):\n",
    "    # Given values\n",
    "    S_A = compute_score(df, model_name)  # The score of Player A\n",
    "    R_B = compute_average(df, model_name)  # The rating of Player \n",
    "\n",
    "    # Define the equation to solve for R_A\n",
    "    def equation(R_A, *data):\n",
    "        S_A, R_B = data\n",
    "        return S_A - 1 / (1 + 10**((R_B - R_A) / K)) \n",
    "\n",
    "    # Solve the equation for R_A\n",
    "    R_A_solution = fsolve(equation, initial_guess, args=(S_A, R_B))\n",
    "\n",
    "    # Extract the calculated R_A value\n",
    "    R_A_calculated = float(R_A_solution[0])\n",
    "    return R_A_calculated\n",
    "\n",
    "computation_Elo(df_non_random.query(\"(white == 'gpt-3.5-turbo-instruct' or black == 'gpt-3.5-turbo-instruct')\")), computation_Elo(df.query('nstarting_move <= 2 and temperature == 0.0')), computation_Elo(df.query('nstarting_move <= 2 and temperature == 0.8')), computation_Elo(df.query('nstarting_move > 2 and temperature == 0.0'))\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "id": "8cec0f29",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1760.0 Elo for gpt-3.5-turbo-instruct against SF and only with legal games/moves\n",
      "1723.0 Elo for gpt-3.5-turbo-instruct against SF and with all games\n",
      "With temperature=0\n",
      "1760.0 Elo for gpt-3.5-turbo-instruct against SF and only with legal games/moves\n",
      "1723.0 Elo for gpt-3.5-turbo-instruct against SF and with all games\n"
     ]
    }
   ],
   "source": [
    "\n",
    "# rs = []\n",
    "# for i in range(0, 1000):\n",
    "#     rs.append(compute_elo(df_non_random.query(\"(white == 'gpt-3.5-turbo-instruct' or black == 'gpt-3.5-turbo-instruct') and `has_illegal` == False\").sample(), candidate_name=\"gpt-3.5-turbo-instruct\", starting_elo=1500, K=400))\n",
    "\n",
    "\n",
    "def lookup_fide_table(score):\n",
    "        # Data here: https://handbook.fide.com/chapter/B022017\n",
    "        data = {\n",
    "            'p': [1.0, .99, .98, .97, .96, .95, .94, .93, .92, .91, .90, .89, .88, .87, .86, .85, .84,\n",
    "                .83, .82, .81, .80, .79, .78, .77, .76, .75, .74, .73, .72, .71, .70, .69, .68, .67,\n",
    "                .66, .65, .64, .63, .62, .61, .60, .59, .58, .57, .56, .55, .54, .53, .52, .51, .50,\n",
    "                .49, .48, .47, .46, .45, .44, .43, .42, .41, .40, .39, .38, .37, .36, .35, .34, .33,\n",
    "                .32, .31, .30, .29, .28, .27, .26, .25, .24, .23, .22, .21, .20, .19, .18, .17, .16,\n",
    "                .15, .14, .13, .12, .11, .10, .09, .08, .07, .06, .05, .04, .03, .02, .01],\n",
    "            'dp': [800, 677, 589, 538, 501, 470, 444, 422, 401, 383, 366, 351, 336, 322, 309, 296, 284,\n",
    "                273, 262, 251, 240, 230, 220, 211, 202, 193, 184, 175, 166, 158, 149, 141, 133, 125,\n",
    "                117, 110, 102, 95, 87, 80, 72, 65, 57, 50, 43, 36, 29, 21, 14, 7, 0, -7, -14, -21, \n",
    "                -29, -36, -43, -50, -57, -65, -72, -80, -87, -95, -102, -110, -117, -125, -133, -141,\n",
    "                -149, -158, -166, -175, -184, -193, -202, -211, -220, -230, -240, -251, -262, -273, \n",
    "                -284, -296, -309, -322, -336, -351, -366, -383, -401, -444, -470, -501, -538, -589, -677, -800]\n",
    "        }\n",
    "\n",
    "        # Convert to DataFrame\n",
    "        df_fide = pd.DataFrame(data)\n",
    "\n",
    "        rscore = round(score, 2)\n",
    "\n",
    "        # lookup score in table\n",
    "        dp = df_fide.loc[df_fide['p'] == rscore, 'dp'].iloc[0]\n",
    "        return dp\n",
    "\n",
    "def fide_elo_computation(dfe, model_name, score=None):\n",
    "    average_opponents_ratings = compute_average(dfe, model_name)\n",
    "    if score is None:\n",
    "        score = compute_score(dfe, model_name)\n",
    "    dp = lookup_fide_table(score)\n",
    "\n",
    "\n",
    "    return average_opponents_ratings + dp\n",
    "\n",
    "                    \n",
    "def compute_elo(df_l, model_gpt_name):\n",
    "    df_elo = df_l.query(f\"(white == '{model_gpt_name}' or black == '{model_gpt_name}') and `has_illegal` == False\")\n",
    "    fide_elo_value = round(fide_elo_computation(df_elo, model_gpt_name), 0)\n",
    "    print(f\"{fide_elo_value} Elo for {model_gpt_name} against SF and only with legal games/moves\")\n",
    "    \n",
    "    df_elo_withillegal = df_l.query(f\"(white == '{model_gpt_name}' or black == '{model_gpt_name}')\")\n",
    "    s, t = compute_score(df_elo_withillegal, model_gpt_name, percentage=False)\n",
    "    sc = s / len(df_elo_withillegal)\n",
    "    fide_elo_with_illegal_value = round(fide_elo_computation(df_elo_withillegal, model_gpt_name, sc), 0)\n",
    "    print(f\"{fide_elo_with_illegal_value} Elo for {model_gpt_name} against SF and with all games\")\n",
    "\n",
    "# Usage:\n",
    "compute_elo(df_non_random, 'gpt-3.5-turbo-instruct')\n",
    "\n",
    "\n",
    "print(\"With temperature=0\")\n",
    "df_non_random['temperature'] = df_non_random['temperature'].astype(float)\n",
    "compute_elo(df_non_random.query(\"temperature == 0.0\"), 'gpt-3.5-turbo-instruct')\n",
    "# print(\"With temperature=0.8\")\n",
    "# compute_elo(df_non_random.query(\"temperature != 0\"), 'gpt-3.5-turbo-instruct')\n",
    "\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "id": "756f697b",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Elo with altered prompt\n",
      "Elo with non altered prompt\n",
      "1760.0 Elo for gpt-3.5-turbo-instruct against SF and only with legal games/moves\n",
      "1723.0 Elo for gpt-3.5-turbo-instruct against SF and with all games\n",
      "Elo with non altered prompt, t=0\n",
      "1760.0 Elo for gpt-3.5-turbo-instruct against SF and only with legal games/moves\n",
      "1723.0 Elo for gpt-3.5-turbo-instruct against SF and with all games\n"
     ]
    }
   ],
   "source": [
    "print(\"Elo with altered prompt\")\n",
    "# compute_elo(df_non_random.query(\"pgn_base.str.contains('Rennes FRA')\"), 'gpt-3.5-turbo-instruct')\n",
    "# compute_elo(df_non_random.query(\"pgn_base.str.contains('Rennes FRA')\").query(\"temperature == 0.0\"), 'gpt-3.5-turbo-instruct')\n",
    "# print(\"we don't have interactions temperature/altered prompt\")\n",
    "# compute_elo(df_non_random.query(\"pgn_base.str.contains('Rennes FRA')\").query(\"temperature == 0.8\"), 'gpt-3.5-turbo-instruct')\n",
    "\n",
    "print(\"Elo with non altered prompt\")\n",
    "compute_elo(df_non_random.query(\"not pgn_base.str.contains('Rennes FRA')\"), 'gpt-3.5-turbo-instruct')\n",
    "print(\"Elo with non altered prompt, t=0\")\n",
    "compute_elo(df_non_random.query(\"not pgn_base.str.contains('Rennes FRA')\").query(\"temperature == 0.0\"), 'gpt-3.5-turbo-instruct')\n",
    "# print(\"Elo with non altered prompt, t=0.8\")\n",
    "# compute_elo(df_non_random.query(\"not pgn_base.str.contains('Rennes FRA')\").query(\"temperature == 0.8\"), 'gpt-3.5-turbo-instruct')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "id": "1b23fa17",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1723.0 Elo for gpt-3.5-turbo-instruct against SF and with all games and only with black pieces\n",
      "1760.0 Elo for gpt-3.5-turbo-instruct against SF and only with legal games/moves and only with black pieces\n"
     ]
    }
   ],
   "source": [
    "# df_elo_onlywhite = df_non_random.query(\"(white == 'gpt-3.5-turbo-instruct') and `has_illegal` == False\")\n",
    "# print(round(fide_elo_computation(df_elo_onlywhite, 'gpt-3.5-turbo-instruct'), 0), \"Elo for gpt-3.5-turbo-instruct against SF and only with legal games/moves and only with white pieces\")\n",
    "\n",
    "# df_elo_onlywhite_withillegal = df_non_random.query(\"(white == 'gpt-3.5-turbo-instruct')\")\n",
    "# mname = 'gpt-3.5-turbo-instruct'\n",
    "# s, t = compute_score(df_elo_onlywhite_withillegal, mname, percentage=False)\n",
    "# sc = s / len(df_elo_onlywhite_withillegal)\n",
    "# print(round(fide_elo_computation(df_elo_onlywhite_withillegal, mname, sc), 0), \"Elo for\", mname, \"against SF and with all games and only with white pieces\")\n",
    "\n",
    "# only black\n",
    "df_elo_onlyblack_withillegal = df_non_random.query(\"(black == 'gpt-3.5-turbo-instruct')\")\n",
    "mname = 'gpt-3.5-turbo-instruct'\n",
    "s, t = compute_score(df_elo_onlyblack_withillegal, mname, percentage=False)\n",
    "sc = s / len(df_elo_onlyblack_withillegal)\n",
    "print(round(fide_elo_computation(df_elo_onlyblack_withillegal, mname, sc), 0), \"Elo for\", mname, \"against SF and with all games and only with black pieces\")\n",
    "\n",
    "# only black\n",
    "df_elo_onlyblack = df_non_random.query(\"(black == 'gpt-3.5-turbo-instruct') and `has_illegal` == False\")\n",
    "print(round(fide_elo_computation(df_elo_onlyblack, 'gpt-3.5-turbo-instruct'), 0), \"Elo for gpt-3.5-turbo-instruct against SF and only with legal games/moves and only with black pieces\")\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "id": "c9c01b2e",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 1080x360 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "def extract_elo_ratings(df_l, model_gpt_name):\n",
    "    wrs = [] # winning games \n",
    "    lrs = [] # losing games\n",
    "    drs = [] # drawing games\n",
    "    \n",
    "    # Filter and iterate over the dataframe\n",
    "    for _, row in df_l.query(f\"(white == '{model_gpt_name}' or black == '{model_gpt_name}') and `has_illegal` == False\").iterrows():\n",
    "        if row['white'] == model_gpt_name and row['result'] == '1-0':\n",
    "            wrs.append(row['black_elo'])\n",
    "        elif row['black'] == model_gpt_name and row['result'] == '0-1':\n",
    "            wrs.append(row['white_elo'])\n",
    "        elif row['black'] == model_gpt_name and row['result'] == '1-0':\n",
    "            lrs.append(row['white_elo'])\n",
    "        elif row['white'] == model_gpt_name and row['result'] == '0-1':\n",
    "            lrs.append(row['black_elo'])\n",
    "        elif row['result'] == '1/2-1/2':\n",
    "            if row['white'] == model_gpt_name:\n",
    "                drs.append(row['black_elo'])\n",
    "            else:\n",
    "                drs.append(row['white_elo'])\n",
    "        else:\n",
    "            print(\"warning\")\n",
    "            print(row)\n",
    "            continue\n",
    "\n",
    "    wrs = [float(val) for val in wrs]\n",
    "    lrs = [float(val) for val in lrs]\n",
    "    drs = [float(val) for val in drs]\n",
    "    \n",
    "    return wrs, lrs, drs\n",
    "\n",
    "def plot_elo_distribution(wrs, lrs, drs, model_gpt_name):\n",
    "    # Compute unique Elos and their counts\n",
    "    unique_elos = sorted(np.unique(wrs))\n",
    "    win_counts = [wrs.count(elo) for elo in unique_elos]\n",
    "    draw_counts = [drs.count(elo) for elo in unique_elos]\n",
    "    lose_counts = [lrs.count(elo) for elo in unique_elos]\n",
    "\n",
    "    # Plotting\n",
    "    bar_width = 0.3\n",
    "    index = np.arange(len(unique_elos))\n",
    "    plt.figure(figsize=(15, 5))\n",
    "    plt.bar(index, win_counts, bar_width, alpha=0.7, color='green', label='Winning Games')\n",
    "    plt.bar(index + bar_width, draw_counts, bar_width, alpha=0.7, color='blue', label='Drawing Games')\n",
    "    plt.bar(index + 2 * bar_width, lose_counts, bar_width, alpha=0.7, color='red', label='Losing Games')\n",
    "\n",
    "    plt.xlabel('Elo Rating')\n",
    "    plt.ylabel('Number of Games')\n",
    "    plt.title(f'Distribution of Elo Ratings by Game Result for {model_gpt_name}')\n",
    "\n",
    "    plt.xticks(index + bar_width, unique_elos, rotation=45)\n",
    "    plt.legend()\n",
    "    plt.tight_layout()\n",
    "    plt.grid(True, which='both', linestyle='--', linewidth=0.5)\n",
    "    plt.savefig(f\"elo_distribution-scores-{model_gpt_name}.png\", dpi=300, bbox_inches='tight')\n",
    "    plt.show()\n",
    "\n",
    "def analyze_and_plot_elo_distribution(df_l, model_gpt_name):\n",
    "    wrs, lrs, drs = extract_elo_ratings(df_l, model_gpt_name)\n",
    "    plot_elo_distribution(wrs, lrs, drs, model_gpt_name)\n",
    "\n",
    "\n",
    "\n",
    "# Return the functions for further use\n",
    "analyze_and_plot_elo_distribution(df_non_random, \"gpt-3.5-turbo-instruct\")\n",
    "\n",
    "# df_non_random\n",
    "\n",
    "\n",
    "# analyze_and_plot_elo_distribution(df_non_random.query(\"not pgn_base.str.contains('Rennes FRA')\"), 'gpt-3.5-turbo-instruct')\n",
    "# analyze_and_plot_elo_distribution(df_non_random.query(\"pgn_base.str.contains('Rennes FRA')\"), 'gpt-3.5-turbo-instruct')\n",
    "\n",
    "\n",
    "\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "id": "f7bfc171",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Final Elo of gpt-3.5-turbo-instruct: 1760.4938271604938\n"
     ]
    }
   ],
   "source": [
    "final_elo = fide_elo_computation(df.query('nstarting_move <= 2 and temperature == 0.0'), 'gpt-3.5-turbo-instruct') # starting_elo=1700, K=40)\n",
    "print(f\"Final Elo of gpt-3.5-turbo-instruct: {final_elo}\")\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "id": "b17adf86",
   "metadata": {},
   "outputs": [],
   "source": [
    "for k, v in df_non_random.query(\"(white == 'gpt-3.5-turbo-instruct' or black == 'gpt-3.5-turbo-instruct') and `has_illegal` == False\").iterrows():\n",
    "    if v['white'] == \"gpt-3.5-turbo-instruct\" and v['black_elo'] != '?':\n",
    "            opponent_elo = float(v['black_elo'])\n",
    "    elif v['black'] == \"gpt-3.5-turbo-instruct\" and v['white_elo'] != '?':\n",
    "        opponent_elo = float(v['white_elo'])\n",
    "    else:\n",
    "        print(\"warning: no elo found for opponent\", k, v['folder_name'])\n",
    "        continue\n",
    "    result = v['result']\n",
    "    if result == '*':\n",
    "        print(\"warning: no result found for opponent\", k, v['folder_name'])\n",
    "        continue"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "id": "93e78377",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0"
      ]
     },
     "execution_count": 40,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(df.query('nstarting_move > 2'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "id": "23faccf9",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>white</th>\n",
       "      <th>black</th>\n",
       "      <th>white_elo</th>\n",
       "      <th>black_elo</th>\n",
       "      <th>result</th>\n",
       "      <th>nmoves</th>\n",
       "      <th>nstarting_move</th>\n",
       "      <th>pgn_base</th>\n",
       "      <th>temperature</th>\n",
       "      <th>random_engine</th>\n",
       "      <th>has_illegal</th>\n",
       "      <th>illegal_move</th>\n",
       "      <th>folder_name</th>\n",
       "      <th>pgn_base_moves</th>\n",
       "      <th>base_pgn_prompt</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "Empty DataFrame\n",
       "Columns: [white, black, white_elo, black_elo, result, nmoves, nstarting_move, pgn_base, temperature, random_engine, has_illegal, illegal_move, folder_name, pgn_base_moves, base_pgn_prompt]\n",
       "Index: []"
      ]
     },
     "execution_count": 41,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df.query('nstarting_move > 2')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "id": "cce4efc3",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([False, True], dtype=object)"
      ]
     },
     "execution_count": 42,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df['has_illegal'].unique()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "id": "5922ffd8",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "       72\n",
       "0-1     9\n",
       "Name: illegal_move, dtype: int64"
      ]
     },
     "execution_count": 43,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df['illegal_move'].value_counts()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "id": "6fa9ed56",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "False    81\n",
       "Name: base_pgn_prompt, dtype: int64"
      ]
     },
     "execution_count": 44,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df['base_pgn_prompt'].value_counts()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "id": "861eb3f7",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "1762.0670615316344"
      ]
     },
     "execution_count": 45,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "computation_Elo(df.query('base_pgn_prompt == False'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "13c76112",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "id": "8f4c9e45",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>white</th>\n",
       "      <th>black</th>\n",
       "      <th>white_elo</th>\n",
       "      <th>black_elo</th>\n",
       "      <th>result</th>\n",
       "      <th>nmoves</th>\n",
       "      <th>nstarting_move</th>\n",
       "      <th>pgn_base</th>\n",
       "      <th>temperature</th>\n",
       "      <th>random_engine</th>\n",
       "      <th>has_illegal</th>\n",
       "      <th>illegal_move</th>\n",
       "      <th>folder_name</th>\n",
       "      <th>pgn_base_moves</th>\n",
       "      <th>base_pgn_prompt</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "Empty DataFrame\n",
       "Columns: [white, black, white_elo, black_elo, result, nmoves, nstarting_move, pgn_base, temperature, random_engine, has_illegal, illegal_move, folder_name, pgn_base_moves, base_pgn_prompt]\n",
       "Index: []"
      ]
     },
     "execution_count": 46,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df.query(\"(white == 'gpt-4' or black == 'gpt-4')\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "id": "edac6af4",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>white</th>\n",
       "      <th>black</th>\n",
       "      <th>white_elo</th>\n",
       "      <th>black_elo</th>\n",
       "      <th>result</th>\n",
       "      <th>nmoves</th>\n",
       "      <th>nstarting_move</th>\n",
       "      <th>pgn_base</th>\n",
       "      <th>temperature</th>\n",
       "      <th>random_engine</th>\n",
       "      <th>has_illegal</th>\n",
       "      <th>illegal_move</th>\n",
       "      <th>folder_name</th>\n",
       "      <th>pgn_base_moves</th>\n",
       "      <th>base_pgn_prompt</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "Empty DataFrame\n",
       "Columns: [white, black, white_elo, black_elo, result, nmoves, nstarting_move, pgn_base, temperature, random_engine, has_illegal, illegal_move, folder_name, pgn_base_moves, base_pgn_prompt]\n",
       "Index: []"
      ]
     },
     "execution_count": 47,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df.query(\"(white == 'gpt-3.5-turbo' or black == 'gpt-3.5-turbo')\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "id": "04183c78",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/mathieuacher/.local/lib/python3.8/site-packages/scipy/optimize/_minpack_py.py:178: RuntimeWarning: The iteration is not making good progress, as measured by the \n",
      "  improvement from the last ten iterations.\n",
      "  warnings.warn(msg, RuntimeWarning)\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "1700.0"
      ]
     },
     "execution_count": 48,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "computation_Elo(df.query(\"(white == 'gpt-4' or black == 'gpt-4') and has_illegal == False\"), model_name='gpt-4')"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.8.19"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}