-
Notifications
You must be signed in to change notification settings - Fork 0
/
astai_dqn.py
233 lines (208 loc) · 10.4 KB
/
astai_dqn.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
import numpy as np
import random
import sys
from collections import deque
import os
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'
import tensorflow as tf
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import GRU, Dense, Dropout, Flatten
from tensorflow.keras.optimizers import Adam
from tensorflow.keras.models import load_model
from tensorflow.keras import backend as K
from stable_baselines3.common.vec_env import DummyVecEnv, SubprocVecEnv
from stable_baselines3.common import logger
from stable_baselines3.common.monitor import Monitor
from price_grabber import get_closing_prices
from get_price_data import get_price_data, get_top_pairs, get_ohlc_data
from stock_trading_env import StockTradingEnv
import json
from dqn_agent import DQNAgent
import gc
import tracemalloc
import gym
from gym import spaces
from gym.utils import seeding
import multiprocessing
import torch
from collections import deque
torch.set_num_threads(1)
def make_env(rank, data, window_size, seed=1, current_step=0):
env = StockTradingEnv(data, window_size, current_step=current_step)
env.seed(seed+rank)
#env = Monitor(env)
return env
def map_values(x, _min, _max):
return (x - _min) / (_max - _min)
def generate_data(prices):
data = get_closing_prices(prices)
# Generate a fake stock price history
close = data['close'].values.tolist()
sma = data['SMA_5'].values.tolist()
rsi = data['RSI_14'].values.tolist()
bbl = data['BBL_3_2.0'].values.tolist()
bbm = data['BBM_3_2.0'].values.tolist()
bbu = data['BBU_3_2.0'].values.tolist()
bbb = data['BBB_3_2.0'].values.tolist()
bbp = data['BBP_3_2.0'].values.tolist()
macd = data['MACD_12_26_9'].values.tolist()
macdh = data['MACDh_12_26_9'].values.tolist()
macds = data['MACDs_12_26_9'].values.tolist()
ema = data['EMA_3'].values.tolist()
print("Loaded Data")
use_set = [close, rsi, bbl, bbm, bbu]
use_set = [[map_values(x, np.nanmin(_set), np.nanmax(_set)) for x in _set] for _set in use_set]
print("Normalized Data")
# Combine the price history and SMA 30 into a 2D array
data = np.array(use_set).T
return data
def write_data(len):
data = {}
pairs = get_top_pairs(len)
for i, pair in enumerate(pairs):
gen_data = generate_data(get_ohlc_data(pair))
if gen_data.size == 3600:
print(f"Generating Data Step {i}")
gen_data = np.array([sub_array for sub_array in gen_data if not np.isnan(sub_array).any()])
data[pair] = gen_data.tolist()
with open("data.json", "w") as outfile:
json.dump(data, outfile, indent = 4)
def update_iteration_data(data):
with open("iteration_data.json", "w") as outfile:
json.dump(data, outfile, indent = 4)
'''
0 - np.array
1 - list
2 - np.float64
3 - bool
4 - list
'''
def write_memories(mem):
memory_dictionary = {'memories': mem}
for _, memories in memory_dictionary.items():
for mem_index, memory in enumerate(memories):
for i, _memory in enumerate(memory):
if type(_memory) == type(np.array([])):
if i == 6:
memory_dictionary[_][mem_index][6] = list(memory_dictionary[_][mem_index][6])
memory_dictionary[_][mem_index][6][0] = list(memory_dictionary[_][mem_index][6][0])
for j, data in enumerate(memory_dictionary[_][mem_index][6][0]):
memory_dictionary[_][mem_index][6][0][j] = float(data)
#print(type(memory_dictionary[_][mem_index][6][0][j]))
#memory_dictionary[_][mem_index][6][0][0] = list
#print(f"Overall {type(memory_dictionary[_][mem_index][6])} Base {memory_dictionary[_][mem_index][6][0]} Type {type(memory_dictionary[_][mem_index][6][0])}")
if i == 0:
memory_dictionary[_][mem_index][0] = list(memory_dictionary[_][mem_index][0])
for entry_index, entry in enumerate(memory_dictionary[_][mem_index][0]):
#print(memory_dictionary[_][mem_index][0])
memory_dictionary[_][mem_index][0][entry_index] = list(memory_dictionary[_][mem_index][0][entry_index])
#print(type(memory_dictionary[_][mem_index][0][entry_index]))
elif type(_memory) == type([]) and i != 0 and i != 6:
for entry_index, entry in enumerate(memory_dictionary[_][mem_index][i]):
#print(entry, i, type(entry))
memory_dictionary[_][mem_index][i][entry_index] = float(_memory[entry_index])
with open("memory.json", "w") as outfile:
json.dump(memory_dictionary, outfile, indent = 4)
print("Memory written to JSON.")
def read_memories():
f = open('memory.json')
memory_dictionary = json.load(f)
_data = deque(maxlen=100)
for _, memories in memory_dictionary.items():
for mem_index, memory in enumerate(memories):
for i, _memory in enumerate(memory):
memory_dictionary[_][mem_index][0] = np.array(memory_dictionary[_][mem_index][0])
memory_dictionary[_][mem_index][2] = [np.float64(memory_dictionary[_][mem_index][2])]
if type(_memory) == type([]) and i == 6:
memory_dictionary[_][mem_index][6] = np.array(memory_dictionary[_][mem_index][6])
memory_dictionary[_][mem_index][6][0] = np.array(memory_dictionary[_][mem_index][6][0])
if type(memory_dictionary[_][mem_index][i]) == type([]):
for entry_index, entry in enumerate(memory_dictionary[_][mem_index][i]):
if i == 0:
memory_dictionary[_][mem_index][i][entry_index] = np.array(entry)
elif i == 2:
memory_dictionary[_][mem_index][i][entry_index] = np.int64(entry[entry_index])
#elif i == 4:
#memory_dictionary[_][mem_index][i][entry_index] = np.float64(entry[entry_index])
try:
memory_dictionary[_][mem_index] = tuple(memory)
except:
print("empty memory")
_data.append(memory_dictionary[_][mem_index])
print("Memories Applied.")
return _data
# Main loop
if __name__ == "__main__":
print("Started")
if not os.path.isfile('data.json'):
write_data(50)
f = open('data.json')
price_data = json.load(f)
print("Loaded Data")
model_name = "main.h5"
window_size = 50
if not os.path.isfile(model_name):
agent = DQNAgent(3, window_size)
sorted_pairs = list(price_data.keys())
episode_start = 0
else:
f = open('iteration_data.json')
data = json.load(f)
agent = DQNAgent(3, window_size, is_model=True, current_iter=data['current_pair'], current_step=data['step'], model_name=model_name, loss=float(data['loss_avg']), epsilon=float(data['epsilon']), learning_rate=float(data['learning_rate']))
[agent.memory.append(memory) for memory in read_memories()]
agent.current_pair = data['current_pair']
sorted_pairs = list(price_data.keys())[agent.current_pair:]
episode_start = int(data['episode'])
print("Loaded Agent")
for i, pair in enumerate(sorted_pairs):
stock_price_data_np = price_data[pair]
num_processes = 1
if not os.path.isfile(model_name):
print("No model found")
envs = make_env(i, stock_price_data_np, window_size)
else:
print("Model Found")
envs = make_env(i, stock_price_data_np, window_size, current_step=agent.step)
print("Loaded Subprocesses")
episodes = 10
batch_size = 32
kill_size = 100
for e in range(episode_start, episodes):
state = envs.reset()
print(pair)
for time in range(1000):
gc.collect()
state = np.array(state[(envs.current_step):window_size+(envs.current_step)])
action, predictions = agent.act(state)
next_state, reward, done, _, info = envs.step(action)
agent.remember(state, action, reward, done, info['n_rewards'], envs.get_best_reward(), predictions)
state = next_state
agent.step = info['step']
agent.current_pair = i
print(f"time_step: {time}, episode: {e}/{episodes}, action: {action}, reward: {np.round(reward, 2)}, net reward: {np.round(info['net reward'], 2)} score: {agent.step}, e: {agent.epsilon}, done: {done}, open orders: {info['orders']}")
print(done)
if done is True or envs.current_step>=(len(stock_price_data_np)-window_size):
print("DONE")
update_iteration_data({"episode": str(e+1), "step": 0, "current_pair": agent.current_iter, "Net Rewards": info['net reward'], "loss": str(agent.loss), "loss_avg": str(agent.loss_avg), "epsilon": str(agent.epsilon), "learning_rate": str(agent.learning_rate)})
kill_size = kill_size - time
break
else:
update_iteration_data({"episode": str(e), "step": info['step'], "current_pair": agent.current_iter, "Net Rewards": info['net reward'], "loss": str(agent.loss), "loss_avg": str(agent.loss_avg), "epsilon": str(agent.epsilon), "learning_rate":str(agent.learning_rate)})
if len(agent.memory) > batch_size:
minibatch = random.sample(agent.memory, batch_size)
agent.replay(minibatch)
agent.save_model(model_name)
write_memories([list(memory) for memory in agent.memory])
if time >= kill_size:
quit()
episode_start = 0
agent.current_iter += 1
agent.learning_rate = agent.learning_rate * agent.learning_rate_decay
f = open('iteration_data.json')
data = json.load(f)
data['episode'] = 0
data['step'] = 0
data['current_pair'] = 0
with open("iteration_data.json", "w") as outfile:
json.dump(data, outfile, indent = 4)
write_data(50)