-
Notifications
You must be signed in to change notification settings - Fork 14
/
Copy pathstage2.py
140 lines (117 loc) · 6.22 KB
/
stage2.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
"""
Stage2: prior learning
run `python stage2.py`
"""
import os
import copy
from argparse import ArgumentParser
import argparse
import torch
import wandb
import numpy as np
import pytorch_lightning as pl
from torch.utils.data import DataLoader
from preprocessing.data_pipeline import build_data_pipeline, build_custom_data_pipeline
from pytorch_lightning.callbacks import LearningRateMonitor
from pytorch_lightning.loggers import WandbLogger
from preprocessing.preprocess_ucr import DatasetImporterUCR, DatasetImporterCustom
from experiments.exp_stage2 import ExpStage2
from evaluation.evaluation import Evaluation
from utils import get_root_dir, load_yaml_param_settings, str2bool
def load_args():
parser = ArgumentParser()
parser.add_argument('--config', type=str, help="Path to the config data file.",
default=get_root_dir().joinpath('configs', 'config.yaml'))
parser.add_argument('--dataset_names', nargs='+', help="e.g., Adiac Wafer Crop`.", default='')
parser.add_argument('--gpu_device_ind', nargs='+', default=[0], type=int, help='Indices of GPU devices to use.')
parser.add_argument('--feature_extractor_type', type=str, default='supervised_fcn', help='rocket | rocket for evaluation.')
parser.add_argument('--use_custom_dataset', type=str2bool, default=False, help='Using a custom dataset, then set it to True.')
return parser.parse_args()
def train_stage2(config: dict,
dataset_name: str,
train_data_loader: DataLoader,
test_data_loader: DataLoader,
gpu_device_ind,
feature_extractor_type:str,
use_custom_dataset:bool,
):
project_name = 'TimeVQVAE-stage2'
# fit
n_classes = len(np.unique(train_data_loader.dataset.Y))
_, in_channels, input_length = train_data_loader.dataset.X.shape
train_exp = ExpStage2(dataset_name, in_channels, input_length, config, n_classes, feature_extractor_type, use_custom_dataset)
n_trainable_params = sum(p.numel() for p in train_exp.parameters() if p.requires_grad)
wandb_logger = WandbLogger(project=project_name, name=None,
config={**config, 'dataset_name': dataset_name, 'n_trainable_params': n_trainable_params, 'feature_extractor_type':feature_extractor_type})
# Check if GPU is available
if not torch.cuda.is_available():
print('GPU is not available.')
# num_cpus = multiprocessing.cpu_count()
num_cpus = 1
print(f'using {num_cpus} CPUs..')
device = num_cpus
accelerator = 'cpu'
else:
accelerator = 'gpu'
device = gpu_device_ind
trainer = pl.Trainer(logger=wandb_logger,
enable_checkpointing=False,
callbacks=[LearningRateMonitor(logging_interval='step')],
max_steps=config['trainer_params']['max_steps']['stage2'],
devices=device,
accelerator=accelerator,
val_check_interval=config['trainer_params']['val_check_interval']['stage2'],
check_val_every_n_epoch=None,
# precision='bf16',
accumulate_grad_batches=1,
)
trainer.fit(train_exp,
train_dataloaders=train_data_loader,
val_dataloaders=test_data_loader
)
print('saving the model...')
if not os.path.isdir(get_root_dir().joinpath('saved_models')):
os.mkdir(get_root_dir().joinpath('saved_models'))
trainer.save_checkpoint(os.path.join(f'saved_models', f'stage2-{dataset_name}.ckpt'))
# test
print('evaluating...')
eval_device = device[0] if accelerator == 'gpu' else 'cpu'
evaluation = Evaluation(dataset_name, in_channels, input_length, n_classes, eval_device, config,
use_neural_mapper=False,
feature_extractor_type=feature_extractor_type,
use_custom_dataset=use_custom_dataset).to(eval_device)
min_num_gen_samples = config['evaluation']['min_num_gen_samples'] # large enough to capture the distribution
(_, _, x_gen), _ = evaluation.sample(max(evaluation.X_test.shape[0], min_num_gen_samples), 'unconditional')
# z_train = evaluation.z_train
z_test = evaluation.z_test
z_gen = evaluation.compute_z_gen(x_gen)
# fid_train = evaluation.fid_score(z_test, z_gen)
wandb.log({'FID': evaluation.fid_score(z_test, z_gen)})
if not use_custom_dataset:
IS_mean, IS_std = evaluation.inception_score(x_gen)
wandb.log({'IS_mean': IS_mean, 'IS_std': IS_std})
# evaluation.log_visual_inspection(evaluation.X_train, x_gen, 'X_train vs X_gen')
evaluation.log_visual_inspection(evaluation.X_test, x_gen, 'X_test vs Xhat')
# evaluation.log_visual_inspection(evaluation.X_train, evaluation.X_test, 'X_train vs X_test')
# evaluation.log_pca([z_train, z_gen], ['z_train', 'z_gen'])
evaluation.log_pca([z_test, z_gen], ['Z_test', 'Zhat'])
# evaluation.log_pca([z_train, z_test], ['z_train', 'z_test'])
wandb.finish()
if __name__ == '__main__':
# load config
args = load_args()
config = load_yaml_param_settings(args.config)
# config
dataset_names = args.dataset_names
# run
for dataset_name in dataset_names:
# data pipeline
batch_size = config['dataset']['batch_sizes']['stage2']
if not args.use_custom_dataset:
dataset_importer = DatasetImporterUCR(dataset_name, **config['dataset'])
train_data_loader, test_data_loader = [build_data_pipeline(batch_size, dataset_importer, config, kind) for kind in ['train', 'test']]
else:
dataset_importer = DatasetImporterCustom(**config['dataset'])
train_data_loader, test_data_loader = [build_custom_data_pipeline(batch_size, dataset_importer, config, kind) for kind in ['train', 'test']]
# train
train_stage2(config, dataset_name, train_data_loader, test_data_loader, args.gpu_device_ind, args.feature_extractor_type, args.use_custom_dataset)