-
Notifications
You must be signed in to change notification settings - Fork 0
/
train.py
171 lines (102 loc) · 4.02 KB
/
train.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
import os
import glob
import argparse
import pickle
from tqdm import tqdm
import IPython
from shutil import copyfile
import gym
import textworld.gym
from textworld import EnvInfos
from time import time
import pandas as pd
import numpy as np
import sys
import yaml
from utils import make_path
from dalab.grid_search.grid_search import GridSearcher
from utils import get_points
from custom_agent import CustomAgent
class Environment:
"""
Wrapper for the TextWorld Environment.
"""
def __init__(self, games_dir, max_nb_steps=100, batch_size=1):
self.games = self.get_games(games_dir)
self.max_nb_steps = max_nb_steps
self.batch_size = batch_size
self.env = self.setup_env()
def step(self, commands):
return self.env.step(commands)
def reset(self):
return self.env.reset()
def render(self):
return self.env.render()
def manual_game(self):
try:
done = False
self.env.reset()
nb_moves = 0
while not done:
self.env.render()
command = input("Input ")
nb_moves += 1
obs, scores, dones, infos = self.env.step([command])
self.env.render() # Final message.
except KeyboardInterrupt:
pass # Quit the game.
print("Played {} steps, scoring {} points.".format(nb_moves, scores[0]))
def setup_env(self):
requested_infos = self.select_additional_infos()
env_id = textworld.gym.register_games(self.games, requested_infos,
max_episode_steps=self.max_nb_steps,
name="training")
env_id = textworld.gym.make_batch(env_id, batch_size=self.batch_size, parallel=True)
return gym.make(env_id)
def get_games(self, games_dir):
games = []
for game in [games_dir]:
if os.path.isdir(game):
games += glob.glob(os.path.join(game, "*.ulx"))
else:
games.append(game)
games = [os.path.join(os.getcwd(), game) for game in games]
print("{} games found for training.".format(len(games)))
return games
def select_additional_infos(self) -> EnvInfos:
request_infos = EnvInfos()
request_infos.description = True
request_infos.inventory = True
request_infos.entities = True
request_infos.verbs = True
request_infos.extras = ["recipe", "walkthrough"]
request_infos.admissible_commands = True
return request_infos
class Trainer:
def __init__(self, game_dir):
self.agent = CustomAgent()
self.env = Environment(game_dir)
def train(self):
self.start_time = time()
for epoch_no in range(1, self.agent.nb_epochs + 1):
for game_no in tqdm(range(len(self.env.games))):
obs, infos = self.env.reset()
self.agent.train()
scores = [0] * len(obs)
dones = [False] * len(obs)
steps = [0] * len(obs)
while not all(dones):
# Increase step counts.
steps = [step + int(not done) for step, done in zip(steps, dones)]
commands = self.agent.act(obs, scores, dones, infos)
obs, scores, dones, infos = self.env.step(commands)
# Let the agent know the game is done.
self.agent.act(obs, scores, dones, infos)
score = sum(scores) / self.agent.batch_size
score, possible_points, percentage = get_points(score, infos['extra.walkthrough'][0])
print('Score: {}/{}'.format(score, possible_points))
if __name__ == '__main__':
parser = argparse.ArgumentParser(description="Train an agent.")
parser.add_argument('--games', action='store', type=str, help='Directory of the games used for training.')
args = parser.parse_args()
Trainer(args.games).train()