-
Notifications
You must be signed in to change notification settings - Fork 23
/
ddm_trainer.py
218 lines (188 loc) · 8.07 KB
/
ddm_trainer.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
import logging
import os
import pathlib
import hydra
import numpy as np
import pandas as pd
from math import floor
from omegaconf import DictConfig, ListConfig, OmegaConf
from sklearn.metrics import r2_score
# TODO: use the model yaml to get the metric
# use the other metrics: MAE and MADE
logger = logging.getLogger("datamodeler")
dir_path = os.path.dirname(os.path.realpath(__file__))
# helper function that return None if element is not present in config
def hydra_read_config_var(cfg: DictConfig, level: str, key_name: str):
"""Reads the config file and returns the config as a dictionary"""
return cfg[level][key_name] if key_name in cfg[level] else None
@hydra.main(config_path="conf", config_name="config")
def main(cfg: DictConfig) -> None:
logger.info("Configuration: ")
logger.info(f"\n{OmegaConf.to_yaml(cfg)}")
# for readability, read common data args into variables
input_cols = hydra_read_config_var(cfg, "data", "inputs")
output_cols = hydra_read_config_var(cfg, "data", "outputs")
augmented_cols = hydra_read_config_var(cfg, "data", "augmented_cols")
iteration_order = hydra_read_config_var(cfg, "data", "iteration_order")
episode_col = hydra_read_config_var(cfg, "data", "episode_col")
iteration_col = hydra_read_config_var(cfg, "data", "iteration_col")
dataset_path = hydra_read_config_var(cfg, "data", "path")
max_rows = hydra_read_config_var(cfg, "data", "max_rows")
test_perc = hydra_read_config_var(cfg, "data", "test_perc")
diff_state = hydra_read_config_var(cfg, "data", "diff_state")
concatenated_steps = hydra_read_config_var(cfg, "data", "concatenated_steps")
concatenated_zero_padding = hydra_read_config_var(
cfg, "data", "concatenated_zero_padding"
)
concatenate_var_length = hydra_read_config_var(cfg, "data", "concatenate_length")
preprocess = hydra_read_config_var(cfg, "data", "preprocess")
var_rename = hydra_read_config_var(cfg, "data", "var_rename")
exogeneous_variables = hydra_read_config_var(cfg, "data", "exogeneous_variables")
exogeneous_save_path = hydra_read_config_var(cfg, "data", "exogeneous_save_path")
initial_values_save_path = hydra_read_config_var(
cfg, "data", "initial_values_save_path"
)
# common model args
save_path = cfg["model"]["saver"]["filename"]
model_name = cfg["model"]["name"]
run_sweep = cfg["model"]["sweep"]["run"]
split_strategy = cfg["model"]["sweep"]["split_strategy"]
results_csv_path = cfg["model"]["sweep"]["results_csv_path"]
ts_model = False
if model_name.lower() == "pytorch":
from all_models import available_models
elif model_name.lower() in ["nhits", "tftmodel", "varima", "ets", "sfarima"]:
from timeseriesclass import darts_models as available_models
ts_model = True
else:
from model_loader import available_models
if ts_model:
from timeseriesclass import TimeSeriesDarts
Model = TimeSeriesDarts
fit_params = cfg["model"]["fit_params"]
else:
Model = available_models[model_name]
fit_params = None
# TODO, decide whether to always save to outputs directory
if cfg["data"]["full_or_relative"] == "relative":
dataset_path = os.path.join(dir_path, dataset_path)
save_path = os.path.join(dir_path, save_path)
if type(input_cols) == ListConfig:
input_cols = list(input_cols)
if type(output_cols) == ListConfig:
output_cols = list(output_cols)
elif type(output_cols) == DictConfig:
output_cols = list(output_cols.keys())
if type(augmented_cols) == ListConfig:
augmented_cols = list(augmented_cols)
model = Model()
# Add extra preprocessing step inside load_csv
# should be done before concatenate_steps
if ts_model:
feature_cols = augmented_cols
label_cols = output_cols
train_df, test_df = model.load_from_csv(
dataset_path,
episode_col,
iteration_col,
label_cols,
feature_cols,
test_perc,
return_ts=False,
var_rename=var_rename,
exogeneous_variables=exogeneous_variables,
exogeneous_path=exogeneous_save_path,
)
else:
X_train, y_train, X_test, y_test = model.load_csv(
dataset_path=dataset_path,
input_cols=input_cols,
augm_cols=augmented_cols,
output_cols=output_cols,
iteration_order=iteration_order,
episode_col=episode_col,
iteration_col=iteration_col,
# drop_nulls: bool = True,
max_rows=max_rows,
test_perc=test_perc,
diff_state=diff_state,
prep_pipeline=preprocess,
var_rename=var_rename,
concatenated_steps=concatenated_steps,
concatenated_zero_padding=concatenated_zero_padding,
concatenate_var_length=concatenate_var_length,
exogeneous_variables=exogeneous_variables,
exogeneous_save_path=exogeneous_save_path,
initial_values_save_path=initial_values_save_path,
)
logger.info(
f"From the full dataset, {test_perc * 100}% will be used for test, while {(1 - test_perc) * 100}% for training/sweeping"
)
# X_train, y_train = model.get_train_set(grouped_per_episode=False)
# X_test, y_test = model.get_test_set(grouped_per_episode=False)
# save training and test sets
save_data_path = os.path.join(os.getcwd(), "data")
if not os.path.exists(save_data_path):
pathlib.Path(save_data_path).mkdir(parents=True, exist_ok=True)
if not ts_model:
logger.info(f"Saving data to {os.path.abspath(save_data_path)}")
np.save(os.path.join(save_data_path, "x_train.npy"), X_train)
np.save(os.path.join(save_data_path, "y_train.npy"), y_train)
np.save(os.path.join(save_data_path, "x_test.npy"), X_test)
np.save(os.path.join(save_data_path, "y_test.npy"), y_test)
logger.info("Building model...")
if ts_model:
model.build_model(
model_type=cfg["model"]["name"],
scale_data=cfg["model"]["scale_data"],
build_params=cfg["model"]["build_params"],
)
else:
model.build_model(**cfg["model"]["build_params"])
if run_sweep:
# TODO: implement sweep for darts class
params = OmegaConf.to_container(cfg["model"]["sweep"]["params"])
logger.info(f"Sweeping with parameters: {params}")
sweep_df = model.sweep(
params=params,
X=X_train,
y=y_train,
search_algorithm=cfg["model"]["sweep"]["search_algorithm"],
num_trials=cfg["model"]["sweep"]["num_trials"],
scoring_func=cfg["model"]["sweep"]["scoring_func"],
results_csv_path=results_csv_path,
splitting_criteria=split_strategy,
)
logger.info(f"Sweep results: {sweep_df}")
else:
logger.info("Fitting model...")
if not ts_model:
model.fit(X_train, y_train)
else:
model.fit(train_df, fit_params)
if not ts_model:
y_pred = model.predict(X_test)
logger.info(f"R^2 score is {r2_score(y_test,y_pred)} for test set.")
logger.info(f"Saving model to {save_path}")
model.save_model(filename=save_path)
## save datasets
pd.DataFrame(X_train, columns=model.feature_cols).to_csv(
os.path.join(save_data_path, "x_train.csv")
)
pd.DataFrame(X_test, columns=model.feature_cols).to_csv(
os.path.join(save_data_path, "x_test.csv")
)
pd.DataFrame(y_train, columns=model.label_cols).to_csv(
os.path.join(save_data_path, "y_train.csv")
)
pd.DataFrame(y_test, columns=model.label_cols).to_csv(
os.path.join(save_data_path, "y_test.csv")
)
else:
y_pred = model.predict(test_df, {"n": 1})
# ts_preds = model.predict(test_df, {"n": 1})
# y_pred = []
# for i in range(len(ts_preds)):
# y_pred.append(ts_preds[i].all_values().flatten())
if __name__ == "__main__":
main()