forked from Aksel488/ECA_RC_RL
-
Notifications
You must be signed in to change notification settings - Fork 0
/
main.py
136 lines (99 loc) · 4.51 KB
/
main.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
import os
if not os.path.isdir('models'):
os.makedirs('models')
if not os.path.isdir('runs'):
os.makedirs('runs')
if not os.path.isdir('screenshots'):
os.makedirs('screenshots')
import gymnasium as gym
import config
import random
import numpy as np
import time
import pickle
import utils
from reservoir import Reservoir
from agent import Agent
EPSILON_DECAY = 0.9975 # 0.9997 for 20_000, 0.9995 for 10_000, 0.999 for 5_000, 0.9975 for 1_000
MIN_EPSILON = 0.001
AGGREGATE_STATS_EVERY = 100 # in episodes, save checkpoint, and print stats
c = 0
e = 1
while e > MIN_EPSILON:
e *= EPSILON_DECAY
c += 1
print(f'Epsilon will reach {MIN_EPSILON} after {c} epochs')
env = gym.make("CartPole-v1")
FILE_NAME = f'runs/{config.RUN_NAME}.txt'
run_start_time = int(time.time())
for rule in utils.unique_eca_rules():
epsilon = 1
MODEL_PATH = f'models/{config.RUN_NAME}/rule{rule}_num{config.NUM}_w{config.WIDTH}_iter{config.ITERATIONS}_input{config.NUM_ROWS_INPUT}_acc{config.ACCURASY_PER_OBSERVATION}_{run_start_time}'
reservoir = Reservoir(rule, render=False)
reservoir.save(f'{MODEL_PATH}/reservoir.pkl')
agent = Agent(reservoir=reservoir)
ep_rewards = []
s = time.time()
t_steps = 0
for episode in range(1, config.EPISODES):
episode_reward = 0
step = 1
observation, info = env.reset()
reservoir.reset()
reservoir.update(observation, env.observation_space)
state = reservoir.read()
early_stop = False
terminated = False
while not terminated:
if random.random() < epsilon:
action = env.action_space.sample()
else:
action = np.argmax(agent.get_actions(state).numpy())
observation, reward, done, truncated, info = env.step(action)
if done: # when terminated, reward is still 1
reward = 0
reservoir.update(observation, env.observation_space)
new_state = reservoir.read()
episode_reward += reward
agent.update_replay_memory((state, action, reward, new_state, done))
agent.train(done, step)
state = new_state
if step >= 200 or done:
terminated = True
step += 1
t_steps += 1
ep_rewards.append(episode_reward)
if not episode % AGGREGATE_STATS_EVERY or episode == 1:
average_reward = sum(ep_rewards[-AGGREGATE_STATS_EVERY:])/len(ep_rewards[-AGGREGATE_STATS_EVERY:])
min_reward = min(ep_rewards[-AGGREGATE_STATS_EVERY:])
max_reward = max(ep_rewards[-AGGREGATE_STATS_EVERY:])
elaps_time_format = time.strftime('%H:%M:%S', time.gmtime(time.time()-s))
print(elaps_time_format, t_steps, episode, episode_reward, max_reward, average_reward, min_reward)
agent.save(f'{MODEL_PATH}/{episode}episode_{max_reward:_>7.2f}max_{average_reward:_>7.2f}avg_{min_reward:_>7.2f}min.pkl')
# if episode > 500 and average_reward > 100:
# run_string = f'Total steps {t_steps:6d}, Rule {rule:3d} stopped after {episode:4d} episodes, finishing EARLY due to GOOD average {average_reward:.2f}, taking {elaps_time_format}'
# utils.write_run_file(FILE_NAME, run_string)
# early_stop = True
# break
data = {'rule': rule, 'rewards': ep_rewards}
with open(f'{MODEL_PATH}/data.pkl', 'wb') as f:
pickle.dump(data, f)
if epsilon > MIN_EPSILON:
epsilon *= EPSILON_DECAY
epsilon = max(MIN_EPSILON, epsilon)
if not early_stop:
average_reward = sum(ep_rewards[-AGGREGATE_STATS_EVERY:])/len(ep_rewards[-AGGREGATE_STATS_EVERY:])
elaps_time_format = time.strftime('%H:%M:%S', time.gmtime(time.time()-s))
run_string = f'Total steps {t_steps:6d}, Rule {rule:3d} stopped after {episode:4d} episodes with an average of average {average_reward:.2f}, taking {elaps_time_format}'
utils.write_run_file(FILE_NAME, run_string)
agent.save(f'{MODEL_PATH}/done.pkl')
data = {'rule': rule, 'rewards': ep_rewards}
with open(f'{MODEL_PATH}/data.pkl', 'wb') as f:
pickle.dump(data, f)
print('-----------------------')
print('-----------------------')
elaps_time = time.time()-s
print(f'Rule {rule:3d}, took {elaps_time:.2f}s, total steps {t_steps}, averaging {t_steps/elaps_time:.2f} steps per second')
print('-----------------------')
print('-----------------------')
env.close()