Skip to content

Latest commit

 

History

History
81 lines (38 loc) · 2.51 KB

5_How_to_use_eval_main.md

File metadata and controls

81 lines (38 loc) · 2.51 KB
Raw

How to use eval main

Settings

At the eval_main.py, you should make a few settings.

simple board example

There are player and enemy for the evaluation.

Player and enemy can be decided as follows.

  • human: human can play the game
  • uct: play Omok with pure MCTS based on UCB
  • puct: play Omok with pure MCTS based on PUCT
  • random: play Omok without any algorithm, it just play with random moves
  • model path: If you write path of the model, it plays Omok with AlphaOmok algorithm with the variables that you set.
  • web: Human can play Omok on the web environment

Also, there is Monitor_model_path for web playing. You should set a model for this, then you can use this to compare you play with AlphaOmok agent. This will be specifically described on the Testing (Web) part.

Parameters

If you want to test saved AlphaOmok model, you should set N_BLOCKS, IN_PLANES and OUT_PLANES according to the model. You can refer to the log file for finding those paramters of the model.

N_MCTS is the number of MCTS iterations. 4 steps of MCTS is 1 iteration. 4 steps of MCTS are as follows.

- Selection

- Expansion

- Simulation (AlphaGo zero doesn't have this process)

- Back up

Therefore, 400 N_MCTS represents 400 iterations of MCTS.

N_MATCH is the number of the games for testing.

Testing (Local)

If you finished the settings with human and run the eval_main.py, you can see the Omok playing between the agents or you can play Omok against the agent with pygame. You can see the demo that I played Omok with the AlphaOmok model as follows.

simple board example

## Testing (Web)

If you finished the setting with web and run eval_main.py, you should open the web browser and go to the following link.

http://127.0.0.1:5000/dashboard

Then you can see the following web page.

simple board example

This is very good for verifying the model and debugging.

Also, if you set the Monitor_model_path, you can compare your play with AlphaOmok agent. After your move is finished, you can check the Player Policy Network and Player MCTS Visit. The output of player policy network represents the probability of each move. Also, the output of player MCTS visit represents the move of the AlphaOmok agent.