main.py

import sys
import os
import gym
import torch
import matplotlib.pyplot as plt
from a3c.discrete_A3C import train
from a3c.utils import v_wrap
from a3c.net import Net
from wordle_env.wordle import WordleEnvBase

def evaluate_checkpoints(dir, env):
    n_s = env.observation_space.shape[0]
    n_a = env.action_space.n
    words_list = env.words
    word_width = len(env.words[0])
    net = Net(n_s, n_a, words_list, word_width)
    results = {}
    print(dir)
    for checkpoint in os.listdir(dir):
            checkpoint_path = os.path.join(dir, checkpoint)
            if os.path.isfile(checkpoint_path):
                net.load_state_dict(torch.load(checkpoint_path))
                wins, guesses = evaluate(net, env)
                results[checkpoint] = wins, guesses
    return dict(sorted(results.items(), key=lambda x: (x[1][0], -x[1][1]), reverse=True))


def evaluate(net, env):
    print("Evaluation mode")
    n_wins = 0
    n_guesses = 0
    n_win_guesses = 0
    env = env.unwrapped
    N = env.allowable_words
    for goal_word in env.words[:N]:
        win, outcomes = play(net, env)
        if win:
            n_wins += 1
            n_win_guesses += len(outcomes)
        # else:
        #     print("Lost!", goal_word, outcomes)
        n_guesses += len(outcomes)
    print(f"Evaluation complete, won {n_wins/N*100}% and took {n_win_guesses/n_wins} guesses per win, "
          f"{n_guesses / N} including losses.")
    return n_wins/N*100, n_win_guesses/n_wins

def play(net, env):
    state = env.reset()
    outcomes = []
    win = False
    for i in range(env.max_turns):
        action = net.choose_action(v_wrap(state[None, :]))
        state, reward, done, _ = env.step(action)
        outcomes.append((env.words[action], reward))
        if done:
            if reward >= 0:
                win = True
            break
    return win, outcomes

def print_results(global_ep, win_ep, res):
    print("Jugadas:", global_ep.value)
    print("Ganadas:", win_ep.value)
    plt.plot(res)
    plt.ylabel('Moving average ep reward')
    plt.xlabel('Step')
    plt.show()


if __name__ == "__main__":
    max_ep = int(sys.argv[1]) if len(sys.argv) > 1 else 100000
    env_id = sys.argv[2] if len(sys.argv) > 2 else 'WordleEnv100FullAction-v0'
    evaluation = True if len(sys.argv) > 3 and sys.argv[3] == 'evaluation' else False
    env = gym.make(env_id)
    model_checkpoint_dir = os.path.join('checkpoints', env.unwrapped.spec.id)
    if not evaluation:
        global_ep, win_ep, gnet, res = train(env, max_ep, model_checkpoint_dir)
        print_results(global_ep, win_ep, res)
        evaluate(gnet, env)
    else:
        results = evaluate_checkpoints(model_checkpoint_dir, env)
        print(results)