-
Notifications
You must be signed in to change notification settings - Fork 21
/
executor.py
182 lines (166 loc) · 10.6 KB
/
executor.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
import shutil
from torch.utils.data import DataLoader
import torch
from loaders.custom_dataset import CustomDataset
from models.DeepLob.deeplob import DeepLOB
from models.iTransformer.itransformer import ITransformer
from models.Transformer.transformer import Transformer
from models.LobTransformer.lobtransformer import LobTransformer
from models.DLA.DLA import DLA
from models.CNN1.cnn1 import CNN1
from models.CNN2.cnn2 import CNN2
from models.AxialLob.axiallob import AxialLOB
from models.TABL.bin_tabl import BiN_BTABL, BiN_CTABL
from optimizers.lightning_batch_gd import BatchGDManager
from loggers import logger
from utils import create_tree, get_training_test_stocks_as_string
class Executor:
def __init__(self, experiment_id, general_hyperparameters, model_hyperparameters, torch_dataset_preparation=False, torch_dataset_preparation_backtest=False):
self.manager = None
self.model = None
self.experiment_id = experiment_id
self.torch_dataset_preparation = torch_dataset_preparation
self.torch_dataset_preparation_backtest = torch_dataset_preparation_backtest
self.training_stocks_string, self.test_stocks_string = get_training_test_stocks_as_string(general_hyperparameters)
if self.torch_dataset_preparation:
create_tree(f"./torch_datasets/threshold_{model_hyperparameters['threshold']}/batch_size_{model_hyperparameters['batch_size']}/training_{self.training_stocks_string}_test_{self.test_stocks_string}/{model_hyperparameters['prediction_horizon']}/")
if general_hyperparameters["model"] == "deeplob":
self.model = DeepLOB(lighten=model_hyperparameters["lighten"])
elif general_hyperparameters["model"] == "transformer":
self.model = Transformer(lighten=model_hyperparameters["lighten"])
elif general_hyperparameters["model"] == "itransformer":
self.model = ITransformer(lighten=model_hyperparameters["lighten"])
elif general_hyperparameters["model"] == "lobtransformer":
self.model = LobTransformer(lighten=model_hyperparameters["lighten"])
elif general_hyperparameters["model"] == "dla":
self.model = DLA(lighten=model_hyperparameters["lighten"])
elif general_hyperparameters["model"] == "cnn1":
self.model = CNN1()
elif general_hyperparameters["model"] == "cnn2":
self.model = CNN2()
elif general_hyperparameters["model"] == "binbtabl":
self.model = BiN_BTABL(120, 40, 100, 5, 3, 1)
elif general_hyperparameters["model"] == "binctabl":
self.model = BiN_CTABL(120, 40, 100, 5, 120, 5, 3, 1)
elif general_hyperparameters["model"] == "axiallob":
self.model = AxialLOB()
if self.torch_dataset_preparation:
# Prepare the training dataloader.
dataset = CustomDataset(
dataset=general_hyperparameters["dataset"],
learning_stage="training",
window_size=model_hyperparameters["history_length"],
shuffling_seed=model_hyperparameters["shuffling_seed"],
cache_size=1,
lighten=model_hyperparameters["lighten"],
threshold=model_hyperparameters["threshold"],
all_horizons=general_hyperparameters["horizons"],
prediction_horizon=model_hyperparameters["prediction_horizon"],
targets_type=general_hyperparameters["targets_type"],
balanced_dataloader=model_hyperparameters["balanced_sampling"],
training_stocks=general_hyperparameters["training_stocks"],
validation_stocks=general_hyperparameters["target_stocks"],
target_stocks=general_hyperparameters["target_stocks"]
)
torch.save(dataset, f"./torch_datasets/threshold_{model_hyperparameters['threshold']}/batch_size_{model_hyperparameters['batch_size']}/training_{self.training_stocks_string}_test_{self.test_stocks_string}/{model_hyperparameters['prediction_horizon']}/training_dataset.pt")
elif self.torch_dataset_preparation is False and self.torch_dataset_preparation_backtest is False:
dataset = torch.load(f"./torch_datasets/threshold_{model_hyperparameters['threshold']}/batch_size_{model_hyperparameters['batch_size']}/training_{self.training_stocks_string}_test_{self.test_stocks_string}/{model_hyperparameters['prediction_horizon']}/training_dataset.pt")
self.train_loader = DataLoader(
dataset,
batch_size=model_hyperparameters["batch_size"],
shuffle=False,
num_workers=model_hyperparameters["num_workers"],
sampler=dataset.glob_indices,
)
if self.torch_dataset_preparation:
# Prepare the validation dataloader.
dataset = CustomDataset(
dataset=general_hyperparameters["dataset"],
learning_stage="validation",
window_size=model_hyperparameters["history_length"],
shuffling_seed=model_hyperparameters["shuffling_seed"],
cache_size=1,
lighten=model_hyperparameters["lighten"],
threshold=model_hyperparameters["threshold"],
all_horizons=general_hyperparameters["horizons"],
targets_type=general_hyperparameters["targets_type"],
prediction_horizon=model_hyperparameters["prediction_horizon"],
training_stocks=general_hyperparameters["training_stocks"],
validation_stocks=general_hyperparameters["target_stocks"],
target_stocks=general_hyperparameters["target_stocks"]
)
torch.save(dataset, f"./torch_datasets/threshold_{model_hyperparameters['threshold']}/batch_size_{model_hyperparameters['batch_size']}/training_{self.training_stocks_string}_test_{self.test_stocks_string}/{model_hyperparameters['prediction_horizon']}/validation_dataset.pt")
elif self.torch_dataset_preparation is False and self.torch_dataset_preparation_backtest is False:
dataset = torch.load(f"./torch_datasets/threshold_{model_hyperparameters['threshold']}/batch_size_{model_hyperparameters['batch_size']}/training_{self.training_stocks_string}_test_{self.test_stocks_string}/{model_hyperparameters['prediction_horizon']}/validation_dataset.pt")
self.val_loader = DataLoader(
dataset,
batch_size=model_hyperparameters["batch_size"],
shuffle=False,
num_workers=model_hyperparameters["num_workers"],
)
if self.torch_dataset_preparation is False and self.torch_dataset_preparation_backtest:
dataset = CustomDataset(
dataset=general_hyperparameters["dataset"],
learning_stage="test",
window_size=model_hyperparameters["history_length"],
shuffling_seed=model_hyperparameters["shuffling_seed"],
cache_size=1,
lighten=model_hyperparameters["lighten"],
threshold=model_hyperparameters["threshold"],
all_horizons=general_hyperparameters["horizons"],
targets_type=general_hyperparameters["targets_type"],
prediction_horizon=model_hyperparameters["prediction_horizon"],
backtest=True,
training_stocks=general_hyperparameters["training_stocks"],
validation_stocks=general_hyperparameters["target_stocks"],
target_stocks=general_hyperparameters["target_stocks"]
)
torch.save(dataset, f"./torch_datasets/threshold_{model_hyperparameters['threshold']}/batch_size_{model_hyperparameters['batch_size']}/training_{self.training_stocks_string}_test_{self.test_stocks_string}/{model_hyperparameters['prediction_horizon']}/test_dataset_backtest.pt")
elif self.torch_dataset_preparation and self.torch_dataset_preparation_backtest is False:
dataset = CustomDataset(
dataset=general_hyperparameters["dataset"],
learning_stage="test",
window_size=model_hyperparameters["history_length"],
shuffling_seed=model_hyperparameters["shuffling_seed"],
cache_size=1,
lighten=model_hyperparameters["lighten"],
threshold=model_hyperparameters["threshold"],
all_horizons=general_hyperparameters["horizons"],
targets_type=general_hyperparameters["targets_type"],
prediction_horizon=model_hyperparameters["prediction_horizon"],
training_stocks=general_hyperparameters["training_stocks"],
validation_stocks=general_hyperparameters["target_stocks"],
target_stocks=general_hyperparameters["target_stocks"]
)
torch.save(dataset, f"./torch_datasets/threshold_{model_hyperparameters['threshold']}/batch_size_{model_hyperparameters['batch_size']}/training_{self.training_stocks_string}_test_{self.test_stocks_string}/{model_hyperparameters['prediction_horizon']}/test_dataset.pt")
elif self.torch_dataset_preparation is False and self.torch_dataset_preparation_backtest is False:
dataset = torch.load(f"./torch_datasets/threshold_{model_hyperparameters['threshold']}/batch_size_{model_hyperparameters['batch_size']}/training_{self.training_stocks_string}_test_{self.test_stocks_string}/{model_hyperparameters['prediction_horizon']}/test_dataset.pt")
self.test_loader = DataLoader(
dataset,
batch_size=model_hyperparameters["batch_size"],
shuffle=False,
num_workers=model_hyperparameters["num_workers"],
)
if self.torch_dataset_preparation is False and self.torch_dataset_preparation_backtest is False:
self.manager = BatchGDManager(
experiment_id=experiment_id,
model=self.model,
train_loader=self.train_loader,
val_loader=self.val_loader,
test_loader=self.test_loader,
epochs=model_hyperparameters["epochs"],
learning_rate=model_hyperparameters["learning_rate"],
patience=model_hyperparameters["patience"],
general_hyperparameters=general_hyperparameters,
model_hyperparameters=model_hyperparameters,
)
def execute_training(self):
self.manager.train()
def execute_testing(self):
self.manager.test()
def logger_clean_up(self):
folder_path = f"{logger.find_save_path(self.experiment_id)}/wandb/"
try:
shutil.rmtree(folder_path)
except:
pass