gen_model.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Generative text model\n",
    "\n",
    "We add a text node to the Bayesian network, conditioned on all diagnoses and symptoms. These conditional text distributions are parameterized by 32 Gaussians, one for each unique combination of diagnosis and symptom values. The rest of the nodes in the Bayesian network are parameterized by regular conditional probability tables which can be learned from the data using maximum likelihood estimation. \n",
    "\n",
    "![title](figures/models_gen.png)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "%load_ext autoreload\n",
    "%autoreload 2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "import pickle\n",
    "with open(\"data/train_4000_final.p\", \"rb\") as file: \n",
    "    train_df = pickle.load(file)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "import pickle\n",
    "with open(\"data/test_1000_final.p\", \"rb\") as file: \n",
    "    test_df = pickle.load(file)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "class_map = {\"season\": {\"warm\": 0, \"cold\": 1}, \n",
    "            \"pneu\": {\"no\": 0, \"yes\": 1}, \"inf\": {\"no\": 0, \"yes\": 1}, \n",
    "            \"dysp\": {\"no\": 0, \"yes\": 1}, \"cough\": {\"no\": 0, \"yes\": 1}, \"nasal\": {\"no\": 0, \"yes\": 1}}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "c:\\Users\\Paloma\\anaconda3\\envs\\bn-text\\Lib\\site-packages\\tqdm\\auto.py:21: TqdmWarning: IProgress not found. Please update jupyter and ipywidgets. See https://ipywidgets.readthedocs.io/en/stable/user_install.html\n",
      "  from .autonotebook import tqdm as notebook_tqdm\n",
      "c:\\Users\\Paloma\\Work\\bn-text\\utils\\data.py:112: UserWarning: Creating a tensor from a list of numpy.ndarrays is extremely slow. Please consider converting the list to a single numpy.ndarray with numpy.array() before converting to a tensor. (Triggered internally at C:\\cb\\pytorch_1000000000000\\work\\torch\\csrc\\utils\\tensor_new.cpp:248.)\n",
      "  preprocessed_data[\"text\"] = torch.tensor(self.dataframe[self.emb_type].tolist(), dtype=torch.float, device=self.device)\n"
     ]
    }
   ],
   "source": [
    "from utils.data import TabDataset\n",
    "import torch\n",
    "\n",
    "emb_type = \"BioLORD emb\"\n",
    "device = torch.device(\"cuda\" if torch.cuda.is_available() else \"cpu\") # put train and test data on the device\n",
    "train_set = TabDataset(train_df, class_map, emb_type, device)\n",
    "test_set = TabDataset(test_df, class_map, emb_type, device)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "In utils.models, you can find the class GenerativeModel. Fitting the distributions in this model consists of two steps.\n",
    "1. learn all conditional probability tables (CPTs), parameterizing the regular portion of the BN: P(season), P(pneu|season), P(inf|season), P(dysp|pneu), P(cough|pneu,inf), P(nasal|inf)\n",
    "2. fit conditional text distributions as Gaussians to parameterize P(text|pneu,inf,dysp,cough,nasal)\n",
    "\n",
    "The code below can be used to do the first step. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "from torch.utils.data import DataLoader\n",
    "from utils.models import GenerativeModel, GenerativeModelAbl\n",
    "from torch.optim import Adam\n",
    "\n",
    "def train_model_generative_CPTs(train, test, node_names, device, model_type=\"gen\", bs_train=100, epochs=100, seed=2023, lr=0.05):\n",
    "    \"\"\"\n",
    "    training loop for fitting CPT portion of generative model\n",
    "    \n",
    "    train: set of training samples\n",
    "    test: set of test samples\n",
    "    node_names: list of names of regular non-text nodes in the BN\n",
    "    device: device to load the model to (GPU/CPU)\n",
    "    model_type: type of model (\"gen\" or \"gen_abl\")\n",
    "    \n",
    "    bs_train: batch size to use for training\n",
    "    epochs: number of epochs to train for\n",
    "    seed: seed to use for initialization\n",
    "    lr: learning rate \n",
    "\n",
    "    returns\n",
    "        train_loss: list of train losses across epochs\n",
    "        test_loss: list of test losses across epochs\n",
    "        train_loss_individual: dict of individual train losses across epochs (i.e. logP for each separate conditional distribution)\n",
    "        test_loss_individual: dict of individual test losses across epochs (i.e. logP for each separate conditional distribution)\n",
    "        model: trained BN\n",
    "    \"\"\"\n",
    "    \n",
    "    torch.manual_seed(seed)\n",
    "    \n",
    "    train_loader = DataLoader(train, batch_size=bs_train, shuffle=True)\n",
    "    test_loader = DataLoader(test, batch_size=len(test), shuffle=False)\n",
    "\n",
    "    # put model on the device\n",
    "    if model_type == \"gen\":\n",
    "        model = GenerativeModel(seed, device)\n",
    "        model.to(device)\n",
    "    elif model_type == \"gen_abl\":\n",
    "        model = GenerativeModelAbl(seed, device)\n",
    "        model.to(device)\n",
    "\n",
    "    adam = Adam(params=model.parameters(), lr=lr)\n",
    "\n",
    "    train_loss_individual = {key:[] for key in node_names}\n",
    "    test_loss_individual = {key:[] for key in node_names}\n",
    "    train_loss = []\n",
    "    test_loss = []\n",
    "\n",
    "    for epoch in range(epochs):\n",
    "\n",
    "        epoch_loss_individual = {key:0 for key in node_names}\n",
    "        epoch_loss = 0\n",
    "\n",
    "        for i, x in enumerate(train_loader): \n",
    "\n",
    "            model.train()\n",
    "            adam.zero_grad()\n",
    "\n",
    "            LL_fully_obs, LL_part_obs, LL_fully_obs_per_var = model(x) # log likelihood of batch under current parametrization of the model\n",
    "            batch_loss = -(LL_part_obs.sum()+LL_fully_obs.sum()) # batch loss is sum of all individual losses \n",
    "            batch_loss.backward()\n",
    "\n",
    "            LL_individual = {key: loss.sum() for key, loss in LL_fully_obs_per_var.items()} # sum individual losses per var\n",
    "            epoch_loss_individual = {key: loss + -LL_individual[key].item() for key, loss in epoch_loss_individual.items()} # store total losses for individual variables per epoch\n",
    "            epoch_loss += batch_loss.item()\n",
    "            \n",
    "            torch.nn.utils.clip_grad_value_(model.parameters(), 5)\n",
    "            adam.step()\n",
    "        \n",
    "        train_loss_individual = {key: loss_list + [epoch_loss_individual[key]] for key, loss_list in train_loss_individual.items()}\n",
    "        train_loss.append(epoch_loss/len(train))\n",
    "            \n",
    "        with torch.no_grad():\n",
    "            model.eval()\n",
    "            for x_test in test_loader: \n",
    "                LL_fully_obs, LL_part_obs, LL_fully_obs_per_var = model(x_test)\n",
    "                LL_individual = {key: loss.sum() for key, loss in LL_fully_obs_per_var.items()} # sum individual losses per var\n",
    "                test_loss_individual = {key: loss_list + [-LL_individual[key].item()] for key, loss_list in test_loss_individual.items()}\n",
    "                test_loss.append(-(LL_fully_obs.sum()+LL_part_obs.sum()).item()/len(test))\n",
    "\n",
    "    for key in train_loss_individual: \n",
    "        train_loss_individual[key] = [e/train.len_sympt_obs() for e in train_loss_individual[key]] # only count the times when this loss was counted = number of records where symptoms are observed\n",
    "        test_loss_individual[key] = [e/test.len_sympt_obs() for e in test_loss_individual[key]]\n",
    "    \n",
    "    return train_loss, test_loss, train_loss_individual, test_loss_individual, model"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The hyperparameters bs_train, epochs and lr were optimized using a separate train/val split. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.model_selection import train_test_split\n",
    "seed = 2024\n",
    "subtrain_df, val_df = train_test_split(train_df, train_size=0.8, random_state=seed) # use val_df in code below to optimize hyperparameters"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Using this train/val split, we landed on the configuration below. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "bs_train = 256\n",
    "epochs = 15\n",
    "lr = 0.05\n",
    "seed = 2024\n",
    "node_names = [\"season\", \"pneu\", \"inf\", \"dysp\", \"cough\", \"nasal\"]\n",
    "\n",
    "train_loss, test_loss, train_loss_individual, test_loss_individual, model = train_model_generative_CPTs(train_set, test_set, node_names, device, bs_train=bs_train, epochs=epochs, seed=seed, lr=lr)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The plot below shows that all invidiual CPT losses have converged. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 864x576 with 6 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "from utils.evaluation import plot_loss\n",
    "plot_loss(train_loss_individual, test_loss_individual, epochs, train_loss, test_loss)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Now we execute step 2, meaning we parameterize the conditional text distribution P(text|pneu,inf,dysp,cough,nasal) for every possible configuration of the set {pneu,inf,dysp,cough,nasal}. We do this based on the subset of training samples where text is actually observed, since there is no need to model empty text embeddings. Hyperparameter alpha was also optimized based on a train-validation split. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_obs = train_df.dropna(axis=0, how=\"any\") # samples where symptoms are not observed cannot contribute to any conditional text distribution, which is conditioned on the value of the symptoms\n",
    "train_text = train_obs[train_obs[\"text\"] != \"\"] # only select fully observed samples where text is not empty "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "c:\\Users\\Paloma\\Work\\bn-text\\utils\\models.py:119: UserWarning: cov(): degrees of freedom is <= 0 (Triggered internally at C:\\cb\\pytorch_1000000000000\\work\\aten\\src\\ATen\\native\\Correlation.cpp:117.)\n",
      "  else:\n"
     ]
    }
   ],
   "source": [
    "alpha = 0.85\n",
    "n_emb = 768\n",
    "model.add_gaussians(train_text, alpha, n_emb)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Finally, we can use the fitted model to predict the following probabilities: \n",
    "- P(Diag=yes|background,symptoms,text)\n",
    "- P(Diag=yes|background,symptoms) \n",
    "- P(Diag=yes|background,text)\n",
    "\n",
    "We use an exact Bayesian inference procedure for this, which is implemented in infer_query and used by the predict_diagnoses function (see utils.evaluation). \n",
    "\n",
    "We also calculate the KL divergence between the learned CPTs and the ground-truth CPTs. \n",
    "\n",
    "The code below fits the generative model for 5 seeds and calculates these outcomes for every seed. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "cpu\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "c:\\Users\\Paloma\\Work\\bn-text\\utils\\models.py:119: UserWarning: cov(): degrees of freedom is <= 0 (Triggered internally at C:\\cb\\pytorch_1000000000000\\work\\aten\\src\\ATen\\native\\Correlation.cpp:117.)\n",
      "  cov = torch.cov(embs.t()) # from docs of cov function: rows are variables, columns are observations -> need to transpose emb\n",
      "c:\\Users\\Paloma\\Work\\bn-text\\utils\\models.py:119: UserWarning: cov(): degrees of freedom is <= 0 (Triggered internally at C:\\cb\\pytorch_1000000000000\\work\\aten\\src\\ATen\\native\\Correlation.cpp:117.)\n",
      "  cov = torch.cov(embs.t()) # from docs of cov function: rows are variables, columns are observations -> need to transpose emb\n",
      "c:\\Users\\Paloma\\Work\\bn-text\\utils\\models.py:119: UserWarning: cov(): degrees of freedom is <= 0 (Triggered internally at C:\\cb\\pytorch_1000000000000\\work\\aten\\src\\ATen\\native\\Correlation.cpp:117.)\n",
      "  cov = torch.cov(embs.t()) # from docs of cov function: rows are variables, columns are observations -> need to transpose emb\n",
      "c:\\Users\\Paloma\\Work\\bn-text\\utils\\models.py:119: UserWarning: cov(): degrees of freedom is <= 0 (Triggered internally at C:\\cb\\pytorch_1000000000000\\work\\aten\\src\\ATen\\native\\Correlation.cpp:117.)\n",
      "  cov = torch.cov(embs.t()) # from docs of cov function: rows are variables, columns are observations -> need to transpose emb\n",
      "c:\\Users\\Paloma\\Work\\bn-text\\utils\\models.py:119: UserWarning: cov(): degrees of freedom is <= 0 (Triggered internally at C:\\cb\\pytorch_1000000000000\\work\\aten\\src\\ATen\\native\\Correlation.cpp:117.)\n",
      "  cov = torch.cov(embs.t()) # from docs of cov function: rows are variables, columns are observations -> need to transpose emb\n"
     ]
    }
   ],
   "source": [
    "from utils.evaluation import CPD_params, predict_diagnoses, performance_metrics\n",
    "\n",
    "class_map = {\"season\": {\"warm\": 0, \"cold\": 1}, \n",
    "            \"pneu\": {\"no\": 0, \"yes\": 1}, \"inf\": {\"no\": 0, \"yes\": 1}, \n",
    "            \"dysp\": {\"no\": 0, \"yes\": 1}, \"cough\": {\"no\": 0, \"yes\": 1}, \"nasal\": {\"no\": 0, \"yes\": 1}}\n",
    "    \n",
    "alpha = 0.85\n",
    "bs_train = 256\n",
    "epochs = 15\n",
    "lr = 0.05\n",
    "n_emb = 768\n",
    "seed = 2024\n",
    "node_names = [\"season\", \"pneu\", \"inf\", \"dysp\", \"cough\", \"nasal\"]\n",
    "emb_type = \"BioLORD emb\"\n",
    "\n",
    "device = torch.device(\"cuda\" if torch.cuda.is_available() else \"cpu\")\n",
    "print(device)\n",
    "\n",
    "seeds = [422, 957, 267, 956, 781]\n",
    "\n",
    "results = {}\n",
    "\n",
    "for seed in seeds:\n",
    "\n",
    "    results[seed] = {}\n",
    "\n",
    "    ## TRAINING ##\n",
    "\n",
    "    train_loss, test_loss, train_loss_individual, test_loss_individual, model = train_model_generative_CPTs(train_set, test_set, node_names, device, bs_train=bs_train, epochs=epochs, seed=seed, lr=lr)\n",
    "    train_obs = train_df.dropna(axis=0, how=\"any\") # samples where symptoms are not observed cannot contribute to any conditional text distribution, which is conditioned on the value of the symptoms\n",
    "    train_text = train_obs[train_obs[\"text\"] != \"\"] # only select fully observed samples where text is not empty \n",
    "    model.add_gaussians(train_text, alpha, n_emb)\n",
    "\n",
    "    ## EVALUATION ##\n",
    "\n",
    "    # KL divergence\n",
    "    _, _, KL = CPD_params(model, \"gen\")\n",
    "    results[seed][f\"KL\"] = KL\n",
    "\n",
    "    # P(pneu=yes|background,symptoms,text)\n",
    "    pred_df = predict_diagnoses(model, test_set, 1, model_type=\"gen\")\n",
    "    _, ap = performance_metrics(pred_df, \"pneu\", model_type=\"gen\", plot=False) \n",
    "    results[seed][f\"P(pneu|season,symptoms,text) test PR\"] = ap\n",
    "    _, ap = performance_metrics(pred_df, \"inf\", model_type=\"gen\", plot=False) \n",
    "    results[seed][f\"P(inf|season,symptoms,text) test PR\"] = ap\n",
    "\n",
    "    # P(pneu=yes|background,symptoms)\n",
    "    pred_df = predict_diagnoses(model, test_set, 1, model_type=\"gen\", excl_text=True)\n",
    "    _, ap = performance_metrics(pred_df, \"pneu\", model_type=\"gen\", plot=False) \n",
    "    results[seed][f\"P(pneu|season,symptoms) test PR\"] = ap\n",
    "    _, ap = performance_metrics(pred_df, \"inf\", model_type=\"gen\", plot=False) \n",
    "    results[seed][f\"P(inf|season,symptoms) test PR\"] = ap\n",
    "\n",
    "    # P(pneu=yes|background,text)\n",
    "    pred_df = predict_diagnoses(model, test_set, 1, model_type=\"gen\", excl_sympt=True)\n",
    "    _, ap = performance_metrics(pred_df, \"pneu\", model_type=\"gen\", plot=False) \n",
    "    results[seed][f\"P(pneu|season,text) test PR\"] = ap\n",
    "    _, ap = performance_metrics(pred_df, \"inf\", model_type=\"gen\", plot=False) \n",
    "    results[seed][f\"P(inf|season,text) test PR\"] = ap"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{422: {'KL': 0.00394699651936337,\n",
       "  'P(pneu|season,symptoms,text) test PR': 0.587030869333911,\n",
       "  'P(inf|season,symptoms,text) test PR': 0.7972179604573827,\n",
       "  'P(pneu|season,symptoms) test PR': 0.09014052651578963,\n",
       "  'P(inf|season,symptoms) test PR': 0.8888777490041548,\n",
       "  'P(pneu|season,text) test PR': 0.44336844001611814,\n",
       "  'P(inf|season,text) test PR': 0.7627988382682274},\n",
       " 957: {'KL': 0.00406827583550339,\n",
       "  'P(pneu|season,symptoms,text) test PR': 0.587030869333911,\n",
       "  'P(inf|season,symptoms,text) test PR': 0.7960016164194188,\n",
       "  'P(pneu|season,symptoms) test PR': 0.08864676229611904,\n",
       "  'P(inf|season,symptoms) test PR': 0.8888777490041548,\n",
       "  'P(pneu|season,text) test PR': 0.44336511872247425,\n",
       "  'P(inf|season,text) test PR': 0.7603235442632299},\n",
       " 267: {'KL': 0.0033924094495796273,\n",
       "  'P(pneu|season,symptoms,text) test PR': 0.5870232961654527,\n",
       "  'P(inf|season,symptoms,text) test PR': 0.7974501031464627,\n",
       "  'P(pneu|season,symptoms) test PR': 0.08864676229611904,\n",
       "  'P(inf|season,symptoms) test PR': 0.8888777490041548,\n",
       "  'P(pneu|season,text) test PR': 0.4433634644791748,\n",
       "  'P(inf|season,text) test PR': 0.762746324937024},\n",
       " 956: {'KL': 0.0046024547260417725,\n",
       "  'P(pneu|season,symptoms,text) test PR': 0.5870283366131065,\n",
       "  'P(inf|season,symptoms,text) test PR': 0.7959459017652315,\n",
       "  'P(pneu|season,symptoms) test PR': 0.09014052651578963,\n",
       "  'P(inf|season,symptoms) test PR': 0.8888777490041548,\n",
       "  'P(pneu|season,text) test PR': 0.4433634644791748,\n",
       "  'P(inf|season,text) test PR': 0.7622881746529163},\n",
       " 781: {'KL': 0.0034746643656465644,\n",
       "  'P(pneu|season,symptoms,text) test PR': 0.5870283366131065,\n",
       "  'P(inf|season,symptoms,text) test PR': 0.7972320464180347,\n",
       "  'P(pneu|season,symptoms) test PR': 0.08864676229611904,\n",
       "  'P(inf|season,symptoms) test PR': 0.8888777490041548,\n",
       "  'P(pneu|season,text) test PR': 0.4433634644791748,\n",
       "  'P(inf|season,text) test PR': 0.7639706439329073}}"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "results"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Ablation of generative text model\n",
    "\n",
    "We remove the arrows from pneu -> text and inf -> text, meaning we only fit 8 Gaussians instead of 32. The adapted model class is called \"GenerativeModelAbl\". \n",
    "\n",
    "Again, we run over 5 different initialization seeds."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "cpu\n"
     ]
    }
   ],
   "source": [
    "from utils.evaluation import CPD_params, predict_diagnoses, performance_metrics\n",
    "\n",
    "class_map = {\"season\": {\"warm\": 0, \"cold\": 1}, \n",
    "            \"pneu\": {\"no\": 0, \"yes\": 1}, \"inf\": {\"no\": 0, \"yes\": 1}, \n",
    "            \"dysp\": {\"no\": 0, \"yes\": 1}, \"cough\": {\"no\": 0, \"yes\": 1}, \"nasal\": {\"no\": 0, \"yes\": 1}}\n",
    "    \n",
    "alpha = 0.85\n",
    "bs_train = 256\n",
    "epochs = 15\n",
    "lr = 0.05\n",
    "n_emb = 768\n",
    "seed = 2024\n",
    "node_names = [\"season\", \"pneu\", \"inf\", \"dysp\", \"cough\", \"nasal\"]\n",
    "emb_type = \"BioLORD emb\"\n",
    "\n",
    "device = torch.device(\"cuda\" if torch.cuda.is_available() else \"cpu\")\n",
    "print(device)\n",
    "\n",
    "test_set = TabDataset(test_df, class_map, emb_type, device)\n",
    "\n",
    "train_sizes = [4000]\n",
    "seeds = [422, 957, 267, 956, 781]\n",
    "\n",
    "results = {}\n",
    "\n",
    "for seed in seeds:\n",
    "\n",
    "    results[seed] = {}\n",
    "\n",
    "    ## TRAINING ##\n",
    "\n",
    "    train_loss, test_loss, train_loss_individual, test_loss_individual, model = train_model_generative_CPTs(train_set, test_set, node_names, device, bs_train=bs_train, epochs=epochs, seed=seed, lr=lr, model_type=\"gen_abl\")\n",
    "    train_obs = train_df.dropna(axis=0, how=\"any\") # samples where symptoms are not observed cannot contribute to any conditional text distribution, which is conditioned on the value of the symptoms\n",
    "    train_text = train_obs[train_obs[\"text\"] != \"\"] # only select fully observed samples where text is not empty \n",
    "    model.add_gaussians(train_text, alpha, n_emb)\n",
    "\n",
    "    ## EVALUATION ##\n",
    "\n",
    "    # KL divergence\n",
    "    _, _, KL = CPD_params(model, \"gen\")\n",
    "    results[seed][f\"KL\"] = KL\n",
    "\n",
    "    # P(pneu=yes|background,symptoms,text)\n",
    "    pred_df = predict_diagnoses(model, test_set, 1, model_type=\"gen\")\n",
    "    _, ap = performance_metrics(pred_df, \"pneu\", model_type=\"gen\", plot=False) \n",
    "    results[seed][f\"P(pneu|season,symptoms,text) test PR\"] = ap\n",
    "    pred_df = predict_diagnoses(model, test_set, 1, model_type=\"gen\")\n",
    "    _, ap = performance_metrics(pred_df, \"inf\", model_type=\"gen\", plot=False) \n",
    "    results[seed][f\"P(inf|season,symptoms,text) test PR\"] = ap\n",
    "\n",
    "    # P(pneu=yes|background,symptoms)\n",
    "    pred_df = predict_diagnoses(model, test_set, 1, model_type=\"gen\", excl_text=True)\n",
    "    _, ap = performance_metrics(pred_df, \"pneu\", model_type=\"gen\", plot=False) \n",
    "    results[seed][f\"P(pneu|season,symptoms) test PR\"] = ap\n",
    "    pred_df = predict_diagnoses(model, test_set, 1, model_type=\"gen\", excl_text=True)\n",
    "    _, ap = performance_metrics(pred_df, \"inf\", model_type=\"gen\", plot=False) \n",
    "    results[seed][f\"P(inf|season,symptoms) test PR\"] = ap\n",
    "\n",
    "    # P(pneu=yes|background,text)\n",
    "    pred_df = predict_diagnoses(model, test_set, 1, model_type=\"gen\", excl_sympt=True)\n",
    "    _, ap = performance_metrics(pred_df, \"pneu\", model_type=\"gen\", plot=False) \n",
    "    results[seed][f\"P(pneu|season,text) test PR\"] = ap\n",
    "    pred_df = predict_diagnoses(model, test_set, 1, model_type=\"gen\", excl_sympt=True)\n",
    "    _, ap = performance_metrics(pred_df, \"inf\", model_type=\"gen\", plot=False) \n",
    "    results[seed][f\"P(inf|season,text) test PR\"] = ap"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{422: {'KL': 0.00394699651936337,\n",
       "  'P(pneu|season,symptoms,text) test PR': 0.09014052651578963,\n",
       "  'P(inf|season,symptoms,text) test PR': 0.8888777490041548,\n",
       "  'P(pneu|season,symptoms) test PR': 0.09014052651578963,\n",
       "  'P(inf|season,symptoms) test PR': 0.8888777490041548,\n",
       "  'P(pneu|season,text) test PR': 0.09420362698454747,\n",
       "  'P(inf|season,text) test PR': 0.8912535567330802},\n",
       " 957: {'KL': 0.00406827583550339,\n",
       "  'P(pneu|season,symptoms,text) test PR': 0.08864676229611904,\n",
       "  'P(inf|season,symptoms,text) test PR': 0.8888777490041548,\n",
       "  'P(pneu|season,symptoms) test PR': 0.08864676229611904,\n",
       "  'P(inf|season,symptoms) test PR': 0.8888777490041548,\n",
       "  'P(pneu|season,text) test PR': 0.09200772797174481,\n",
       "  'P(inf|season,text) test PR': 0.8913817961646368},\n",
       " 267: {'KL': 0.0033924094495796273,\n",
       "  'P(pneu|season,symptoms,text) test PR': 0.08864676229611904,\n",
       "  'P(inf|season,symptoms,text) test PR': 0.8888777490041548,\n",
       "  'P(pneu|season,symptoms) test PR': 0.08864676229611904,\n",
       "  'P(inf|season,symptoms) test PR': 0.8888777490041548,\n",
       "  'P(pneu|season,text) test PR': 0.09437815026596214,\n",
       "  'P(inf|season,text) test PR': 0.8911309340860604},\n",
       " 956: {'KL': 0.0046024547260417725,\n",
       "  'P(pneu|season,symptoms,text) test PR': 0.09014052651578963,\n",
       "  'P(inf|season,symptoms,text) test PR': 0.8888777490041548,\n",
       "  'P(pneu|season,symptoms) test PR': 0.09014052651578963,\n",
       "  'P(inf|season,symptoms) test PR': 0.8888777490041548,\n",
       "  'P(pneu|season,text) test PR': 0.09262458651055755,\n",
       "  'P(inf|season,text) test PR': 0.8914944344516109},\n",
       " 781: {'KL': 0.0034746643656465644,\n",
       "  'P(pneu|season,symptoms,text) test PR': 0.08864676229611904,\n",
       "  'P(inf|season,symptoms,text) test PR': 0.8888777490041548,\n",
       "  'P(pneu|season,symptoms) test PR': 0.08864676229611904,\n",
       "  'P(inf|season,symptoms) test PR': 0.8888777490041548,\n",
       "  'P(pneu|season,text) test PR': 0.09344697264963092,\n",
       "  'P(inf|season,text) test PR': 0.8917483148222299}}"
      ]
     },
     "execution_count": 16,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "results"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "bn-text",
   "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": "3.11.4"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}