main.py

import os
import sys
import gym
import matplotlib.pyplot as plt
import torch.multiprocessing as mp

from a3c.discrete_A3C import Net, Worker
from a3c.shared_adam import SharedAdam
from a3c.utils import v_wrap
from wordle_env.wordle import WordleEnvBase

os.environ["OMP_NUM_THREADS"] = "1"

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.")

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

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'
    env = gym.make(env_id)
    n_s = env.observation_space.shape[0]
    n_a = env.action_space.n
    words_list = env.words
    word_width = len(env.words[0])
    gnet = Net(n_s, n_a, words_list, word_width)        # global network
    gnet.share_memory()         # share the global parameters in multiprocessing
    opt = SharedAdam(gnet.parameters(), lr=1e-4, betas=(0.92, 0.999))      # global optimizer
    global_ep, global_ep_r, res_queue, win_ep = mp.Value('i', 0), mp.Value('d', 0.), mp.Queue(), mp.Value('i', 0)

    # parallel training
    workers = [Worker(max_ep, gnet, opt, global_ep, global_ep_r, res_queue, i, env, n_s, n_a, words_list, word_width, win_ep) for i in range(mp.cpu_count())]
    [w.start() for w in workers]
    res = []                    # record episode reward to plot
    while True:
        r = res_queue.get()
        if r is not None:
            res.append(r)
        else:
            break
    [w.join() for w in workers]
    print("Jugadas:", global_ep.value)
    print("Ganadas:", win_ep.value)
    plt.plot(res)
    plt.ylabel('Moving average ep reward')
    plt.xlabel('Step')
    plt.show()
    evaluate(gnet, env)