-
Notifications
You must be signed in to change notification settings - Fork 3
/
Copy pathdeepnetwork.py
140 lines (113 loc) · 5.73 KB
/
deepnetwork.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
import tensorflow as tf
import numpy as np
from config import *
import random
import os
import datetime
import csv
class DeepQnetwork:
def __init__(self, training_mode):
self.epsilon, self.min_epsilon = 1, 0.1
self.decay = self.epsilon/((generations//2)-1)
self.training_mode = training_mode
self.optimizer = tf.keras.optimizers.Adam(learning_rate=0.00025)
self.train_network = self.build_network()
self.predict_network = self.build_network()
self.counter=0
'''def get_data(self):
random_choice=random.randrange(len(input_val))
return np.array([input_val[random_choice]]),np.array([one_hot_op[random_choice]])'''
def record_summary(self, _loss, counter):
summary = tf.Summary(value=[tf.Summary.Value(tag="loss", simple_value=_loss)])
self.writer.add_summary(summary, counter)
'''def save_model(self):
self.saver.save(self.sess, os.path.join(*[".","checkpoints","rl_weights"]),global_step=1000)'''
def load_model(self):
latest_weights = tf.train.latest_checkpoint(save_dir)
self.predict_network.load_weights(latest_weights)
def update_prediction_network(self):
for train_grad, pred_grad in zip(self.train_network.trainable_variables, self.train_network.trainable_variables):
pred_grad.assign(train_grad)
self.train_network.save_weights(save_path.format(self.counter))
print("leveling up")
def update_q_value(self, rewards, current_q_list, next_q_list, actions, done):
current_q_list = current_q_list.numpy()
next_max_qs = np.max(next_q_list, axis=1)
new_qs = rewards + (np.ones(done.shape)-done)*discount_factor * next_max_qs
for i in range(len(current_q_list)):
current_q_list[i, actions[i]] = new_qs[i]
return current_q_list
def loss(self, ground_truth, prediction):
loss = tf.keras.losses.mean_squared_error(ground_truth, prediction)
return loss
def get_action(self, state):
if np.random.random() > self.epsilon:
_action = self.get_prediction(np.expand_dims(state, axis=0))
action = np.argmax(_action)
else:
action = np.random.randint(0, no_of_actions)
return action
def get_prediction(self, states):
states = np.reshape(states, newshape=(states.shape[0], IM_H, IM_W, 4))/255
prediction = self.predict_network(states)
return prediction
def predict(self, states):
states = np.reshape(states, newshape=(states.shape[0], IM_H, IM_W, 4))/255
prediction = self.train_network(states)
return prediction
def save_log(self, step, quantity, filename):
with open(os.path.join(log_save_path, filename), 'a+') as fi:
csv_w = csv.writer(fi, delimiter=',')
csv_w.writerow([step, quantity])
@tf.function
def train_step(self, states, actions):
with tf.GradientTape() as tape:
predictions = self.train_network(states)
loss = self.loss(actions, predictions)
gradients = tape.gradient(loss, self.train_network.trainable_variables)
gradients = [tf.clip_by_norm(gradient, 10) for gradient in gradients]
self.optimizer.apply_gradients(zip(gradients, self.train_network.trainable_variables))
return loss
def build_network(self):
'''inp = tf.keras.layers.Input((5,5))
x = tf.keras.layers.Flatten()(inp)
x = tf.keras.layers.Dense(256, activation='relu')(x)
x = tf.keras.layers.Dense(512, activation='relu')(x)
x = tf.keras.layers.Dense(512, activation='relu')(x)
x = tf.keras.layers.Dense(no_of_actions, activation='linear')(x)'''
inp = tf.keras.layers.Input((IM_H, IM_W, 4))
x = tf.keras.layers.Conv2D(32, (3,3), activation='relu')(inp)
x = tf.keras.layers.Conv2D(32, (3,3), activation='relu')(x)
x = tf.keras.layers.Conv2D(64, (3,3), activation='relu')(x)
x = tf.keras.layers.MaxPool2D((2,2))(x)
x = tf.keras.layers.Dropout(0.2)(x)
x = tf.keras.layers.Conv2D(64, (3,3), activation='relu')(x)
x = tf.keras.layers.Conv2D(128, (3,3), activation='relu')(x)
x = tf.keras.layers.MaxPool2D((2,2))(x)
x = tf.keras.layers.Dropout(0.2)(x)
x = tf.keras.layers.Conv2D(128, (3,3), activation='relu')(x)
x = tf.keras.layers.Conv2D(256, (3,3), activation='relu')(x)
x = tf.keras.layers.MaxPool2D((2,2))(x)
x = tf.keras.layers.Dropout(0.2)(x)
x = tf.keras.layers.Conv2D(512, (3,3), activation='relu')(x)
x = tf.keras.layers.MaxPool2D((2,2))(x)
x = tf.keras.layers.Flatten()(x)
x = tf.keras.layers.Dense(512, activation='relu')(x)
x = tf.keras.layers.Dropout(0.1)(x)
x = tf.keras.layers.Dense(512, activation='relu')(x)
x = tf.keras.layers.Dense(no_of_actions, activation='linear')(x)
model = tf.keras.Model(inputs=inp, outputs=x)
model.summary()
return model
def train(self, previous_memories):
self.counter += 1
current_nodes, actions, next_nodes, rewards, done = previous_memories
#print(current_nodes.shape, actions.shape, rewards.shape, next_nodes.shape)
current_action_qs = self.predict(current_nodes)
next_action_qs = self.get_prediction(next_nodes)
current_action_qs = self.update_q_value(rewards, current_action_qs, next_action_qs, actions, done)
current_nodes = np.reshape(current_nodes, newshape=(batch_size, IM_H, IM_W, 4))/255
loss = self.train_step(current_nodes, current_action_qs)
with tf_writer.as_default():
tf.summary.scalar("loss", data=np.mean(loss), step=self.counter)
self.save_log(self.counter, np.mean(loss), "loss.csv")