-
Notifications
You must be signed in to change notification settings - Fork 0
/
training_mode.py
249 lines (208 loc) · 8.65 KB
/
training_mode.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
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
import argparse
import os
import shutil
from random import random, randint, sample
import pygame
import button
import matplotlib
import matplotlib.backends.backend_agg as agg
import pylab
import numpy as np
import torch
import torch.nn as nn
from torch.utils.tensorboard import SummaryWriter
from deep_q_learning import DeepQNetwork
from tetris_cheater import Tetris as Cheater
from tetris_fair import Tetris as Fair
from collections import deque
matplotlib.use("Agg")
def get_args():
parser = argparse.ArgumentParser(
"""Implementation of Deep Q Network to play Tetris""")
parser.add_argument("--batch_size", type=int, default=512)
parser.add_argument("--lr", type=float, default=5e-4)
parser.add_argument("--gamma", type=float, default=0.99)
parser.add_argument("--initial_epsilon", type=float, default=1)
parser.add_argument("--final_epsilon", type=float, default=5e-4)
parser.add_argument("--saved_path", type=str, default="trained_models")
parser.add_argument("--log_path", type=str, default="tensorboard")
parser.add_argument("--save_interval", type=int, default=1000)
parser.add_argument("--num_decay_epochs", type=float, default=2000)
parser.add_argument("--num_epochs", type=int, default=3000)
parser.add_argument("--replay_memory_size", type=int, default=30000)
args = parser.parse_args()
return args
def train(opt, training_type, number_of_features):
# Checks if the device has a supported gpu otherwise use cpu
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
if torch.cuda.is_available():
torch.cuda.manual_seed(123)
else:
torch.manual_seed(123)
if os.path.isdir(opt.log_path):
shutil.rmtree(opt.log_path)
# Creates a folder for the given log path
os.makedirs(opt.log_path)
writer = SummaryWriter(opt.log_path)
# Set and refresh screen
screen = pygame.display.set_mode((1400, 700))
screen.fill((0, 0, 0))
# Modes
if training_type == "fair":
env = Fair(screen, "train", True)
else:
env = Cheater(screen, "train", True)
# model is the neural network
model = DeepQNetwork(number_of_features).to(device)
# Optimises the model using the learning rate
optimizer = torch.optim.Adam(model.parameters(), lr=opt.lr)
# measures the mean squared error between elements
criterion = nn.MSELoss()
# Gets the default states of the environment
state = env.reset().to(device)
# Limits amount of moves made
steps = 0
max_step = 2000
font_small = pygame.font.SysFont('Arial', 20)
clock = pygame.time.Clock()
# Setups abound queue to the length of the reply memory size
replay_memory = deque(maxlen=opt.replay_memory_size)
epoch = 0
score = []
return_button = button.Button((61, 97, 128), 575, 625, 200, 50, 'Return')
pygame.display.flip()
while epoch < opt.num_epochs:
next_steps = env.get_next_states()
# Decides to do exploration or exploitation
epsilon = opt.final_epsilon + (max(opt.num_decay_epochs - epoch, 0) * (
opt.initial_epsilon - opt.final_epsilon) / opt.num_decay_epochs)
u = random()
random_action = u <= epsilon
next_actions, next_states = zip(*next_steps.items())
next_states = torch.stack(next_states).to(device)
# Evaluates model
model.eval()
with torch.no_grad():
predictions = model(next_states)[:, 0]
# Trains model
model.train()
if random_action:
index = randint(0, len(next_steps) - 1)
else:
index = torch.argmax(predictions).item()
next_state = next_states[index, :].to(device)
action = next_actions[index]
# Gets next steps from environment
reward, done = env.step(action)
steps = steps + 1
replay_memory.append([state, reward, next_state, done])
for event in pygame.event.get():
if event.type == pygame.QUIT:
pygame.display.quit()
quit()
pos = pygame.mouse.get_pos()
if event.type == pygame.MOUSEBUTTONDOWN:
if return_button.is_over(pos):
writer.close()
return True
if event.type == pygame.MOUSEMOTION:
if return_button.is_over(pos):
return_button.color = (61, 97, 128)
else:
return_button.color = (147, 150, 153)
area = pygame.Rect(0, 75, 900, 625)
return_button.draw(screen)
fps = font_small.render("fps:" + str(int(clock.get_fps())), True, pygame.Color('white'))
screen.blit(fps, (10, 75))
clock.tick(200)
pygame.display.update(area)
if done or (max_step <= steps):
final_score = env.score
final_pieces_placed = env.total_pieces_placed
final_cleared_lines = env.total_lines_cleared
state = env.reset().to(device)
steps = 0
else:
state = next_state
continue
if len(replay_memory) < opt.replay_memory_size / 10:
continue
score.append(final_score)
epoch += 1
batch = sample(replay_memory, min(len(replay_memory), opt.batch_size))
state_batch, reward_batch, next_state_batch, done_batch = zip(*batch)
state_batch = torch.stack(tuple(state for state in state_batch)).to(device)
reward_batch = torch.from_numpy(np.array(reward_batch, dtype=np.float32)[:, None]).to(device)
next_state_batch = torch.stack(tuple(state for state in next_state_batch)).to(device)
q_values = model(state_batch)
model.eval()
with torch.no_grad():
next_prediction_batch = model(next_state_batch)
model.train()
y_batch = torch.cat(
tuple(reward if done else reward + opt.gamma * prediction for reward, done, prediction in
zip(reward_batch, done_batch, next_prediction_batch)))[:, None]
optimizer.zero_grad()
loss = criterion(q_values, y_batch)
loss.backward()
optimizer.step()
graph_results(score, opt.num_epochs)
writer.add_scalar('Train/Score', final_score, epoch - 1)
writer.add_scalar('Train/Tetrominoes', final_pieces_placed, epoch - 1)
writer.add_scalar('Train/Cleared lines', final_cleared_lines, epoch - 1)
if epoch > 0 and epoch % opt.save_interval == 0:
torch.save(model, "{}/{}_tetris_{}".format(opt.saved_path, training_type, epoch))
torch.save(model, "{}/{}_tetris".format(opt.saved_path, training_type))
writer.close()
display(screen)
# Draws graph
def graph_results(score, length):
fig = pylab.figure(figsize=[4, 4], dpi=90)
ax = fig.gca()
ax.plot(score)
ax.set_title("Agents score over {}/{} Iteration".format(length, len(score)))
ax.set_xlabel('Iteration')
ax.set_ylabel('Score')
canvas = agg.FigureCanvasAgg(fig)
canvas.draw()
renderer = canvas.get_renderer()
raw_data = renderer.tostring_rgb()
pygame.init()
screen = pygame.display.get_surface()
size = canvas.get_width_height()
surf = pygame.image.fromstring(raw_data, size, "RGB")
screen.blit(surf, (800, 200))
area = pygame.Rect(800, 0, 600, 700)
pygame.display.update(area)
pylab.close('all')
# Draws notice at the end of training
def display(screen):
pygame.draw.rect(screen, (71, 73, 74), (1400 / 2 - 200, 200, 400, 300), 0)
selection_menu_button = button.Button((61, 97, 128), 525, 400, 350, 50, 'Selection Menu')
draw_text_middle("Training Complete", 40, (255, 255, 255), screen)
pygame.display.update()
while True:
for event in pygame.event.get():
if event.type == pygame.QUIT:
pygame.display.quit()
quit()
pos = pygame.mouse.get_pos()
if event.type == pygame.MOUSEBUTTONDOWN:
if selection_menu_button.is_over(pos):
return False
if event.type == pygame.MOUSEMOTION:
if selection_menu_button.is_over(pos):
selection_menu_button.color = (61, 97, 128)
else:
selection_menu_button.color = (147, 150, 153)
selection_menu_button.draw(screen)
pygame.display.update()
# Draws centred text
def draw_text_middle(text, size, color, screen):
font = pygame.font.SysFont('Arial', size, bold=True)
label = font.render(text, 1, color)
screen.blit(label, (1400 / 2 - (label.get_width() / 2), 250 - label.get_height() / 2))
def main(training_type, number_of_features):
opt = get_args()
train(opt, training_type, number_of_features)
return True