-
Notifications
You must be signed in to change notification settings - Fork 1
/
Tensorader.py
156 lines (140 loc) · 5.3 KB
/
Tensorader.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
import tensorflow as tf
import math
import numpy as np
lastTrades = []
def doAction(btc, euro, action, price):
"""
print('Action: ', action)
print(' Price: ', price)
print(' €', euro, ' - ', btc)
"""
#TransactionCost = 0.025
TransactionCost = 0.05
if action == 0:
#print('buying @ ', price)
factor = 1
buy = euro * factor / price
btc += buy
euro -= euro * factor
btc -= TransactionCost * buy
"""
if action == 1:
#print('buying @ ', price)
factor = 0.5
buy = euro * factor / price
btc += buy
euro -= euro * factor
euro -= 0.02 * buy
if action == 2:
#print('buying @ ', price)
factor = 0.3
buy = euro * factor / price
btc += buy
euro -= euro * factor
euro -= 0.02 * buy
#if action == 3:
#print('holding on')
if action == 4:
#print('selling ', btc, ' BTC @ ', price)
factor = 0.3
sell = btc * factor * price
euro += sell
btc -= btc * factor
euro -= 0.02 * sell
if action == 5:
#print('selling ', btc, ' BTC @ ', price)
factor = 0.5
sell = btc * factor * price
euro += sell
btc -= btc * factor
euro -= 0.02 * sell
"""
if action == 2:
#print('selling ', btc, ' BTC @ ', price)
factor = 1
sell = btc * factor * price
euro += sell
btc -= btc * factor
euro -= TransactionCost * sell
euro = (math.floor(euro * 100)) / 100.0
return btc, euro
def create_file_reader_ops(filename_queue):
reader = tf.TextLineReader(skip_header_lines=0)
record_defaults = [[0], [0.0], [0.0]]
if len(lastTrades) > 0:
lastTrades.pop(0)
while len(lastTrades) < 50:
_, csv_row = reader.read(filename_queue)
timestamp, price, volume = tf.decode_csv(csv_row, record_defaults=record_defaults)
#lastTrades.append([price, volume])
lastTrades.append([price])
return lastTrades
filename_queue = tf.train.string_input_producer(["..\..\data\.coinbaseEUR.csv"])
trade_reader = create_file_reader_ops(filename_queue)
NumberOfActions = 3
trades_placeholder = tf.placeholder(shape=[None, 50,1],dtype=tf.float32, name='trades')
wallet_placeholder = tf.placeholder(shape=[None, 2], dtype=tf.float32, name='wallet')
actions_placeholder = tf.placeholder(shape=[None], dtype=tf.int32)
rewards_placeholder = tf.placeholder(shape=[None], dtype=tf.float32)
#Y0 = tf.layers.dense(tick_placeholder. tf.reshape(trades_placeholder, [-1, 100]))
Y1 = tf.layers.dense(tf.reshape(trades_placeholder ,[-1,50]), 200, activation=tf.nn.sigmoid)
Y2 = tf.layers.dense(Y1, 150, activation=tf.nn.sigmoid)
Y3 = tf.layers.dense(Y2, 100, activation=tf.nn.sigmoid)
#Y4 = tf.layers.dense(Y3, 100, activation=tf.nn.sigmoid)
#Y5 = tf.layers.dense(Y4, 50, activation=tf.nn.sigmoid)
YN = tf.layers.dense(Y3, NumberOfActions, activation=tf.nn.softmax)
sample_op = tf.multinomial(logits=YN, num_samples=1)
#cross_entropies = tf.losses.softmax_cross_entropy(onehot_labels=tf.one_hot(actions_placeholder, 5), logits=YN)
#loss = tf.reduce_sum(-rewards_placeholder * cross_entropies)
#loss = tf.sigmoid(rewards_placeholder * cross_entropies)
#optimizer = tf.train.RMSPropOptimizer(learning_rate=0.003, decay=0.99)
cross_entropies = tf.losses.softmax_cross_entropy(onehot_labels=tf.one_hot(actions_placeholder, NumberOfActions), logits=YN)
#loss = -(tf.log(cross_entropies)*rewards_placeholder)
loss = tf.log(cross_entropies)*rewards_placeholder
optimizer = tf.train.GradientDescentOptimizer(learning_rate=0.003)
#optimizer = tf.train.AdamOptimizer(learning_rate=0.01)
update = optimizer.minimize(loss)
train_op = optimizer.minimize(loss)
with tf.Session() as sess:
# Start populating the filename queue.
tf.global_variables_initializer().run()
coord = tf.train.Coordinator()
threads = tf.train.start_queue_runners(coord=coord)
reward = -1
i = 0
ticks = 50
last_rewards = []
median = 0
while median < 10:
btc = 0.0
euro = 100.0
observations = []
actions = []
wallets = []
interactions = [0] * NumberOfActions
for tick in range(ticks):
trade_data = sess.run(trade_reader)
price = trade_data[-1][0]
wallet = [euro, btc]
sample_data = {trades_placeholder: [trade_data], wallet_placeholder: [wallet] }
action = sess.run(sample_op, feed_dict=sample_data)
interaction = action[0][0]
observations.append(trade_data)
actions.append(action[0][0])
wallets.append(wallet)
interactions[interaction] = interactions[interaction] + 1
btc, euro = doAction(btc, euro, action[0][0], price)
reward = (((euro + btc * price) / 100) - 1) * 100
#reward = ((euro / 100) - 1) * 100
#rewards.append(reward)
rewards = [reward] * ticks
last_rewards.append(reward)
if i % 25 == 0 :
median = np.median(last_rewards)
print(i,': reward:', reward, '[median: ', median, '] (' , interactions,') €', euro, ' BTC ', btc)
last_rewards.clear()
train_data = { trades_placeholder: observations, wallet_placeholder: wallets, actions_placeholder: actions, rewards_placeholder: rewards}
sess.run(train_op, feed_dict=train_data)
i += 1
coord.request_stop()
coord.join(threads)