import os

import time

import torch

import env

from functions import compare_model

# simulate 면 1초마다 보여줌, 선턴 후턴 다르게 한번씩

is_simulation = False

# agent class 할당, 여기도 모델 파일에 따라 자동화할 수 있는 방법이 없을까?

agents = {}

agents[1] = (

env.MinimaxDQNAgent(eps=0, model_num=6)

)

agents[2] = (

env.MinimaxDQNAgent(eps=0.3, model_num=6)

)

# models for comparison folder 에서 모델 이름 가져오기

# 폴더 안에는 model 파일만 두개가 존재해야함

# 바꿔야하는 경로를 지정합니다.

folder_path = "model/model_for_comparison/"

file_names = os.listdir(folder_path)

for i in range(len(file_names)):

file_name = file_names[i]

if not '.pt' in file_name:

print("it is impossible")

exit()

else:

agents[i+1].policy_net.load_state_dict(torch.load(folder_path+file_name))

agents[i+1].update_target_net()

print("agent {} model name is {}".format(i+1, file_name))

if is_simulation:

for first_player in [1,2]:

CFenv = env.ConnectFourEnv(first_player=first_player)

while CFenv.win == 0:

time.sleep(1)

turn = CFenv.player

state_ = env.board_normalization(False,CFenv, agents[turn].policy_net.model_type)

state = torch.from_numpy(state_).float()

action = agents[turn].select_action(state, CFenv, player=turn)

if isinstance(action, tuple): action = action[0]

CFenv.step(action)

CFenv.print_board(clear_board=False)

else:

record = compare_model(agents[1], agents[2], n_battle=100)

print("win rate of agent 1: {}%".format(record[0]/sum(record)*100))