-
Notifications
You must be signed in to change notification settings - Fork 3
/
main.py
237 lines (210 loc) · 12.8 KB
/
main.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
234
235
236
237
import pandas as pd
import no_transfer_linear, no_transfer_lstm
import global_linear_linear, global_linear_lstm, global_lstm_linear, global_lstm_lstm
import torch
import utils
import pickle
import numpy as np
import random
# data params
manualSeed = 999999999
np.random.seed(manualSeed)
random.seed(manualSeed)
torch.manual_seed(manualSeed)
torch.backends.cudnn.deterministic = True
torch.backends.cudnn.benchmark = False
data_config = {"data_path": ".\\Tasks\\",
"region": ["Americas", "Europe", "Asia and Pacific", "MEA"],
"Europe": ["Europe_AEX", "Europe_ASE", "Europe_ATX", "Europe_BEL20", "Europe_BUX",
"Europe_BVLX", "Europe_CAC", "Europe_CYSMMAPA", "Europe_DAX", "Europe_HEX",
"Europe_IBEX", "Europe_ISEQ", "Europe_KFX", "Europe_OBX", "Europe_OMX",
"Europe_SMI", "Europe_UKX", "Europe_VILSE", "Europe_WIG20", "Europe_XU100",
"Europe_SOFIX", "Europe_SBITOP", "Europe_PX", "Europe_CRO"],
"Asia and Pacific": ["Asia and Pacific_AS51", "Asia and Pacific_FBMKLCI", "Asia and Pacific_HSI",
"Asia and Pacific_JCI", "Asia and Pacific_KOSPI", "Asia and Pacific_KSE100",
"Asia and Pacific_NIFTY", "Asia and Pacific_NKY", "Asia and Pacific_NZSE50FG",
"Asia and Pacific_PCOMP", "Asia and Pacific_STI", "Asia and Pacific_SHSZ300",
"Asia and Pacific_TWSE"],
"Americas": ["Americas_IBOV", "Americas_MEXBOL", "Americas_MERVAL", "Americas_SPTSX", "Americas_SPX",
"Americas_RTY"],
"MEA": ["MEA_DFMGI", "MEA_DSM", "MEA_EGX30", "MEA_FTN098", "MEA_JOSMGNFF",
"MEA_KNSMIDX", "MEA_KWSEPM", "MEA_MOSENEW", "MEA_MSM30", "MEA_NGSE30", "MEA_PASISI",
"MEA_SASEIDX", "MEA_SEMDEX", "MEA_TA-35", "MEA_TOP40"],
"additional_data_path": "_all_assets_data.pkl.gz"
}
# problem params
problem_config = {"export_path": ".\\",
"val_period": 0, # if val is 0, then its results are the same as training
"holdout_period": 252 * 3,
}
# model params
model_config = {"tsteps": 10,
"tasks_tsteps": 0, # [len(data_config[x]) for x in data_config["region"]] - right
"batch_size": 32,
"seq_len": 252,
"transfer_strat": "global_lstm_lstm",
"device": torch.device("cuda"),
"export_losses": False,
"no_transfer_linear": {"opt_lr": 0.001,
"amsgrad": True,
"export_weights": False
},
"no_transfer_lstm": {"opt_lr": 0.001,
"amsgrad": True,
"export_model": False,
"out_nhi": 50,
"nlayers": 2,
"drop_rate": 0.1
},
"global_linear_linear": {"opt_lr": 0.01,
"amsgrad": True,
"export_weights": False,
"in_transfer_dim": 200,
"out_transfer_dim": 200
},
"global_lstm_linear": {"opt_lr": 0.01,
"amsgrad": True,
"export_model": False,
"in_transfer_dim": 5,
"out_transfer_dim": 5,
"nlayers": 2,
"drop_rate": 0.1
},
"global_linear_lstm": {"opt_lr": 0.01,
"amsgrad": True,
"export_model": False,
"in_transfer_dim": 5,
"out_transfer_dim": 5,
"in_nlayers": 2,
"out_nlayers": 2,
"out_nhi": 10,
"drop_rate": 0.1
},
"global_lstm_lstm": {"opt_lr": 0.01,
"amsgrad": True,
"export_model": False,
"in_transfer_dim": 5,
"out_transfer_dim": 5,
"in_nlayers": 2,
"out_nlayers": 2,
"nlayers": 2,
"out_nhi": 10,
"drop_rate": 0.1,
"drop_rate_transfer": 0.1
}
}
# main routine
# pre-allocation
export_label = "valperiod_" + str(problem_config["val_period"]) + "_testperiod_" + str(problem_config["holdout_period"]) + \
"_tsteps_" + str(model_config["tsteps"]) + "_tksteps_" + str(model_config["tasks_tsteps"]) + "_batchsize_" + \
str(model_config["batch_size"]) + "_seqlen_" + str(model_config["seq_len"]) + "_transferstrat_" + \
model_config["transfer_strat"] + "_lr_" + str(model_config[model_config["transfer_strat"]]["opt_lr"])
data_config["export_label"] = export_label
problem_config["export_label"] = export_label
model_config["export_label"] = export_label
model_config["export_path"] = problem_config["export_path"]
# get data
Xtrain_tasks, Xval_tasks, Xtest_tasks = utils.get_data(data_config, problem_config, model_config)
# set model
if model_config["transfer_strat"] == "no_transfer_linear":
transfer_trad_strat = no_transfer_linear.NoTransferLinear(Xtrain_tasks, model_config)
add_label = [""] * len(data_config["region"])
elif model_config["transfer_strat"] == "no_transfer_lstm":
transfer_trad_strat = no_transfer_lstm.NoTransferLSTM(Xtrain_tasks, model_config)
add_label = ["_nhi_" + str(model_config["no_transfer_lstm"]["out_nhi"]) +
"_nlayers_" + str(model_config["no_transfer_lstm"]["nlayers"]) +
"dpr" + str(model_config["no_transfer_lstm"]["drop_rate"]) for x in data_config["region"]]
elif model_config["transfer_strat"] == "global_linear_linear":
transfer_trad_strat = global_linear_linear.GlobalLinearLinear(Xtrain_tasks, model_config)
add_label = ["_indim_" + str(model_config["global_linear_linear"]["in_transfer_dim"]) +
"_outdim_" + str(model_config["global_linear_linear"]["out_transfer_dim"]) for x in data_config["region"]]
elif model_config["transfer_strat"] == "global_lstm_linear":
transfer_trad_strat = global_lstm_linear.GlobalLSTMLinear(Xtrain_tasks, model_config)
add_label = ["_indim_" + str(model_config["global_lstm_linear"]["in_transfer_dim"]) +
"_outdim_" + str(model_config["global_lstm_linear"]["out_transfer_dim"]) +
"_inlay_" + str(model_config["global_lstm_linear"]["nlayers"]) +
"dpr" + str(model_config["global_lstm_linear"]["drop_rate"]) for x in data_config["region"]]
elif model_config["transfer_strat"] == "global_linear_lstm":
transfer_trad_strat = global_linear_lstm.GlobalLinearLSTM(Xtrain_tasks, model_config)
add_label = ["_indim_" + str(model_config["global_linear_lstm"]["in_transfer_dim"]) +
"_outdim_" + str(model_config["global_linear_lstm"]["out_transfer_dim"]) +
"_inlay_" + str(model_config["global_linear_lstm"]["in_nlayers"]) +
"_outlay_" + str(model_config["global_linear_lstm"]["out_nlayers"]) +
"_lindim_" + str(model_config["global_linear_lstm"]["out_nhi"]) +
"dpr" + str(model_config["global_linear_lstm"]["drop_rate"]) for x in data_config["region"]]
elif model_config["transfer_strat"] == "global_lstm_lstm":
transfer_trad_strat = global_lstm_lstm.GlobalLSTMLSTM(Xtrain_tasks, model_config)
add_label = ["_indim_" + str(model_config["global_lstm_lstm"]["in_transfer_dim"]) +
"_outdim_" + str(model_config["global_lstm_lstm"]["out_transfer_dim"]) +
"_inlay_" + str(model_config["global_lstm_lstm"]["in_nlayers"]) +
"_outlay_" + str(model_config["global_lstm_lstm"]["out_nlayers"]) +
"_odim_" + str(model_config["global_lstm_lstm"]["out_nhi"]) +
"_ltr_" + str(model_config["global_lstm_lstm"]["nlayers"]) +
"_dpr_" + str(model_config["global_lstm_lstm"]["drop_rate"]) +
"_dtr_" + str(model_config["global_lstm_lstm"]["drop_rate_transfer"]) for x in data_config["region"]]
# additional labelling
to_add_label = {}
for (lab, region) in zip(add_label, data_config["region"]):
to_add_label[region] = lab
# train model
import time
start=time.time()
transfer_trad_strat.train()
print(time.time()-start)
# get signals
Xtrain_signal = transfer_trad_strat.predict(Xtrain_tasks)
Xval_signal = transfer_trad_strat.predict(Xval_tasks)
Xtest_signal = transfer_trad_strat.predict(Xtest_tasks)
# compute results
k = True
for region in data_config["region"]:
region_task_paths = [t + "_all_assets_data.pkl.gz" for t in data_config[region]]
z = True
for (tk, tk_path) in zip(data_config[region], region_task_paths):
# get signal
pred_train = Xtrain_signal[region][tk].cpu()
pred_val = Xval_signal[region][tk].cpu()
pred_test = Xtest_signal[region][tk].cpu()
# get target
Ytrain = Xtrain_tasks[region][tk].view(1, -1, Xtrain_tasks[region][tk].size(1))[:, 1:].cpu()
Yval = Xval_tasks[region][tk].view(1, -1, Xval_tasks[region][tk].size(1))[:, 1:].cpu()
Ytest = Xtest_tasks[region][tk].view(1, -1, Xtest_tasks[region][tk].size(1))[:, 1:].cpu()
# compute returns
df_train_ret = pred_train.mul(Ytrain)[0].cpu().numpy() - utils.calc_tcosts(pred_train)[0].cpu().numpy()
df_val_ret = pred_val.mul(Yval)[0].cpu().numpy() - utils.calc_tcosts(pred_val)[0].cpu().numpy()
df_test_ret = pred_test.mul(Ytest)[0].cpu().numpy() - utils.calc_tcosts(pred_test)[0].cpu().numpy()
# get performance metrics
df = pd.read_pickle(data_config["data_path"] + tk_path)
df_train_ret = pd.DataFrame(df_train_ret, columns=df.columns)
df_train_metrics = utils.compute_performance_metrics(df_train_ret)
df_train_metrics["exchange"] = tk
df_val_ret = pd.DataFrame(df_val_ret, columns=df.columns)
df_val_metrics = utils.compute_performance_metrics(df_val_ret)
df_val_metrics["exchange"] = tk
df_test_ret = pd.DataFrame(df_test_ret, columns=df.columns)
df_test_metrics = utils.compute_performance_metrics(df_test_ret)
df_test_metrics["exchange"] = tk
if z:
all_df_train_metrics = df_train_metrics.copy()
all_df_val_metrics = df_val_metrics.copy()
all_df_test_metrics = df_test_metrics.copy()
z = False
else:
all_df_train_metrics = pd.concat([all_df_train_metrics, df_train_metrics], axis=0)
all_df_val_metrics = pd.concat([all_df_val_metrics, df_val_metrics], axis=0)
all_df_test_metrics = pd.concat([all_df_test_metrics, df_test_metrics], axis=0)
# export results
all_df_train_metrics["region"] = region
all_df_train_metrics["set"] = "train"
all_df_val_metrics["region"] = region
all_df_val_metrics["set"] = "val"
all_df_test_metrics["region"] = region
all_df_test_metrics["set"] = "test"
pd.concat([all_df_train_metrics, all_df_val_metrics, all_df_test_metrics], axis=0).to_csv(
problem_config["export_path"] + region + "_" + problem_config["export_label"] + to_add_label[region] + ".csv")
pickle.dump(model_config, open(problem_config["export_path"] + region + "_" + problem_config["export_label"] +
to_add_label[region] + "_modelconfig.pkl.gz", "wb"))
if model_config["export_losses"]:
pickle.dump(transfer_trad_strat.losses[region], open(problem_config["export_path"] + region + "_" +
problem_config["export_label"] + to_add_label[region] +
"_losses.pkl.gz", "wb"))