-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathrl_mercher.py
164 lines (132 loc) · 4.59 KB
/
rl_mercher.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
#@title tradingenv
import pandas as pd
import numpy as np
import torch
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
import sys
from ge_env import StockTradingEnv
from loop_helpers import *
from td_agent import TDAgent
from dqn_agent import DQNAgent
def trainAgent():
#CONFIGURABLE PARAMS
decay = .997
min_rand = .1
training_episodes = 1500
model_load_name = ""
target_model_load_name = ""
args = sys.argv
print(args)
agent_type = args[2]
model_save_name = args[3]
if len(args) > 4:
model_load_name = args[4]
if len(args) > 5:
target_model_load_name = args[5]
itemDF = pd.read_csv('./item_data.csv', index_col=0)
itemDF = itemDF[itemDF['id'] == 2]
itemDF = itemDF.drop('dt_updated',axis=1)
bestSoFarModel = None
bestSoFar = 0
learning = []
cumulativeReward = []
cumulativeRewardAverage = []
agent = None
environment = None
if agent_type.upper() == "TD":
agent = TDAgent(decay, min_rand)
environment = StockTradingEnv(itemDF, 9)
elif agent_type.upper() == "DQN":
agent = None
environment = StockTradingEnv(itemDF, 9)
agent = DQNAgent(decay, min_rand)
if agent_type.upper() == "TD" and model_load_name != "":
agent.model.load_state_dict(torch.load(f"./{model_load_name}.pth"))
if agent_type.upper() == "DQN" and model_load_name != "":
agent.policy_net.load_state_dict(torch.load(f"./{model_load_name}.pth"))
agent.target_net.load_state_dict(torch.load(f"./{target_model_load_name}.pth"))
runavgtotal = 0;
#Train Model
for i in range(training_episodes):
if i%20==0:
cumulativeRewardAverage.append(runavgtotal/20)
runavgtotal = 0
print(i)
prof, bestSoFar, bestSoFarModel = learningEpisode(agent_type, agent, environment, cumulativeReward, learning, bestSoFar, bestSoFarModel)
runavgtotal += cumulativeReward[-1]
print(agent.randomness)
#Save models
if agent_type.upper() == "TD":
torch.save(agent.model.state_dict(), f"./{model_save_name}.pth")
torch.save(bestSoFarModel, f"./best-{model_save_name}.pth")
if agent_type.upper() == "DQN":
torch.save(agent.policy_net.state_dict(), f"./{model_save_name}.pth")
torch.save(agent.target_net.state_dict(), f"./{model_save_name}-targ.pth")
mod, targ= bestSoFarModel
torch.save(mod, f"./best-{model_save_name}.pth")
torch.save(targ, f"./best-{model_save_name}-targ.pth")
#test onlastpart
max_steps = 37000
environment = StockTradingEnv(itemDF, 9, max_steps)
agent.randomness = 0
profits = []
gainz = evalLoopEnd(agent, environment)
profits.append(gainz)
x = np.arange(0, len(cumulativeReward), 1)
sns.lineplot(x=x, y=cumulativeReward)
plt.title("reward")
plt.xlabel("Episodes")
plt.ylabel("profit")
plt.show()
x = np.arange(0, len(cumulativeRewardAverage), 1)
sns.lineplot(x=x, y=cumulativeRewardAverage)
plt.title("average reward from 20 episodes")
plt.xlabel("set of 20 Episodes")
plt.ylabel("profit")
plt.show()
def evalAgent():
model_load_name = ""
target_model_load_name = ""
args = sys.argv
print(args)
agent_type = args[2]
if len(args) > 3:
model_load_name = args[3]
if len(args) > 4:
target_model_load_name = args[4]
itemDF = pd.read_csv('./item_data.csv', index_col=0)
itemDF = itemDF[itemDF['id'] == 2]
itemDF = itemDF.drop('dt_updated',axis=1)
agent = None
environment = None
if agent_type.upper() == "TD":
agent = TDAgent(.997, .1)
environment = StockTradingEnv(itemDF, 9)
elif agent_type.upper() == "DQN":
agent = None
environment = StockTradingEnv(itemDF, 15)
agent = DQNAgent(0.997, 0.1, True)
if agent_type.upper() == "TD" and model_load_name != "":
agent.model.load_state_dict(torch.load(f"./{model_load_name}.pth"))
if agent_type.upper() == "DQN" and model_load_name != "":
agent.policy_net.load_state_dict(torch.load(f"./{model_load_name}.pth"))
agent.target_net.load_state_dict(torch.load(f"./{target_model_load_name}.pth"))
#test onlastpart
max_steps = 37000
environment = StockTradingEnv(itemDF, 9, max_steps)
agent.randomness = 0
profits = []
gainz = evalLoopEnd(agent, environment)
profits.append(gainz)
def main():
args = sys.argv
print(args)
runtype = args[1]
if runtype.upper() == "TRAIN":
trainAgent()
elif runtype.upper() == "EVAL":
evalAgent()
if __name__ == "__main__":
main()