forked from DustinSands/Hanabi
-
Notifications
You must be signed in to change notification settings - Fork 0
/
tests.py
133 lines (126 loc) · 5.18 KB
/
tests.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
# -*- coding: utf-8 -*-
"""
Created on Sat Jun 6 09:00:33 2020
@author: Racehorse
"""
import random
import numpy as np
import tensorflow as tf
import helper_functions
from wrapper import wrapper
def test_accelerated():
random.seed(0)
np.random.seed(0)
tf.random.set_seed(0)
ins = wrapper(batch_size = 64,
policy_type = 'epsilon_decay',
policy_param ={'eps':0, 'eps_decay':0.9999, 'min_eps':0.002},
accelerated = True, Double_DQN_version = 0, optimizer = 'adadelta',
# games_per_epoch = 1,
hidden_layers =[100,100,100,100,100])
ins.train(1)
experience = ins.player[0].get_memory_batch(1)
from_obs, ard, to_obs = zip(*experience)
action = ard[0][0]
reward = ard[0][1]
done = ard[0][2]
state = np.array(from_obs)
state2 = np.array(to_obs)
# print([state, np.array([[action,reward,done]]), state2], [0])
loss = ins.training_model.evaluate([state, np.array([[action,reward,done]]), state2],
np.zeros(1))
assert 0.05058867111802101==loss, "Invalid loss for GPU acceleration"
ins = wrapper(batch_size = 64,
policy_type = 'epsilon_decay',
policy_param ={'eps':0, 'eps_decay':0.9999, 'min_eps':0.002},
accelerated = True, Double_DQN_version = 0, optimizer = 'adadelta',
# games_per_epoch = 1,
hidden_layers =[100,100,100,100,100])
ins.train(1)
loss = ins.training_model.evaluate([state, np.array([[action,reward,done]]), state2],
np.zeros(1))
assert 0.09761060774326324==loss, "Invalid loss for GPU acceleration DDQN 1"
ins = wrapper(batch_size = 64,
policy_type = 'epsilon_decay',
policy_param ={'eps':0, 'eps_decay':0.9999, 'min_eps':0.002},
accelerated = True, Double_DQN_version = 0, optimizer = 'adadelta',
# games_per_epoch = 1,
hidden_layers =[100,100,100,100,100])
ins.train(1)
loss = ins.training_model.evaluate([state, np.array([[action,reward,done]]), state2],
np.zeros(1))
assert 0.0347185917198658==loss, "Invalid loss for GPU acceleration DDQN 2"
print('Test passed!')
def compare_accelerated(seed, DoDQN, epochs, done = 2):
"""Compare the performance of the accelerated and normal models.
Trains both models for number of epochs, picks a experience tuple, and prints
Q values as well as loss function for debug purposes. Both models should give
same results."""
global experience
helper_functions.set_seed(seed)
ins = wrapper(batch_size = 1,
policy_type = 'epsilon_decay',
policy_param ={'eps':0, 'eps_decay':0.9999, 'min_eps':0.002},
accelerated = True, Double_DQN_version = DoDQN, optimizer = 'adadelta',
# games_per_epoch = 1,
hidden_layers =[100,100,100,100,100])
ins.train(epochs)
if epochs > 0:
experience = ins.player[0].get_memory_batch(1)
from_obs, ard, to_obs = zip(*experience)
action = ard[0][0]
reward = ard[0][1]
done = ard[0][2]
state = np.array(from_obs)
state2 = np.array(to_obs)
print(f'Done {done}, Action {action}, Reward {reward}')
else:
from_obs, _, _, _ = ins.env.reset()
to_obs, _, _, _ = ins.env.reset()
state = np.array([from_obs])
state2 = np.array([to_obs])
action = 0
reward = 2
print(ins.online_model.predict(state))
print(ins.online_model.predict(state2))
# print([state, np.array([[action,reward,done]]), state2], [0])
loss = ins.training_model.evaluate([state, np.array([[action,reward,done]]), state2],
np.zeros(1))
print(loss)
helper_functions.set_seed(seed)
ins = wrapper(batch_size = 1,
policy_type = 'epsilon_decay',
policy_param ={'eps':0, 'eps_decay':0.9999, 'min_eps':0.002},
accelerated = False, Double_DQN_version = DoDQN, optimizer = 'adadelta',
# games_per_epoch = 1,
hidden_layers =[100,100,100,100,100])
ins.train(epochs)
print(ins.online_model.predict(state))
print(ins.online_model.predict(state2))
from_obs_array = state
to_obs_array = state2
alpha = 1
gamma = 0.9
if DoDQN != 1:
online_next_Q_values = np.array(ins.online_model.predict_on_batch(to_obs_array))
if DoDQN > 0:
target_Q_values = np.array(ins.target_model.predict_on_batch(to_obs_array))
Q_values = np.array(ins.online_model.predict_on_batch(from_obs_array))
if DoDQN==2:
argmax = np.expand_dims(np.argmax(online_next_Q_values, axis = 1), axis = -1)
max_for_next_obs = np.take_along_axis(target_Q_values, argmax, axis = 1)
elif DoDQN == 1:
max_for_next_obs = np.amax(target_Q_values, axis = 1)
else:
max_for_next_obs = np.amax(online_next_Q_values, axis = 1)
Q_values[0][action] *= 1-alpha
if done:
Q_values[0][action] = reward
else:
calc_action_value = alpha*(reward + gamma*max_for_next_obs[0])
Q_values[0][action]+=calc_action_value
# print(from_obs_array, Q_values)
loss = ins.online_model.evaluate(from_obs_array, Q_values)
print(loss)
if __name__ == '__main__':
test_accelerated()