-
Notifications
You must be signed in to change notification settings - Fork 0
/
train_exp1.py
229 lines (188 loc) · 8.6 KB
/
train_exp1.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
from pathlib import Path
from argparse import ArgumentParser
import torch
from torch.utils.data import DataLoader
import pytorch_lightning as pl
from pytorch_lightning.loggers import WandbLogger
from pytorch_lightning.callbacks import ModelCheckpoint, EarlyStopping
try:
import wandb
except ImportError:
print('wandb not available')
try:
import ray
except ImportError:
print('ray not available')
import data
import losses
import models
import ap
def cmdline_args():
parser = ArgumentParser()
# experiment config
parser.add_argument('--project', default=None)
parser.add_argument('--name', default='default')
# dataset config
parser.add_argument('--loss', choices=['hungarian_l2', 'hungarian_ce', 'hungarian_nl'], default='hungarian_l2')
parser.add_argument('--set_size', type=int, default=64)
parser.add_argument('--set_dim', type=int, default=64)
parser.add_argument('--input_dim', type=int, default=4)
parser.add_argument('--dataset_size', type=int, default=64000)
parser.add_argument('--n_obj_per_sample', type=int, default=4)
parser.add_argument('--rand_perm', action='store_true', default=False)
# training config
parser.add_argument('--seed', type=int, default=42)
parser.add_argument('--batch_size', type=int, default=128)
parser.add_argument('--lr', type=float, default=1e-3)
parser.add_argument('--epochs', type=int, default=50)
parser.add_argument('--checkpoint_path', default='checkpoints')
parser.add_argument('--num_data_workers', type=int, default=0)
parser.add_argument('--num_ray_workers', type=int, default=0)
parser.add_argument('--check_val_every_n_epoch', type=int, default=1)
# model config
parser.add_argument('--model', default='idspn', choices=['idspn', 'dspn', 'deepsets', 'lstm', 'transformer_with_pe', 'transformer_no_pe','transformer_rnd_pe'])
parser.add_argument('--latent_dim', type=int, default=64)
parser.add_argument('--hidden_dim', type=int, default=64)
# idspn config
parser.add_argument('--decoder_lr', type=float, default=1.0)
parser.add_argument('--decoder_iters', type=int, default=20)
parser.add_argument('--decoder_momentum', type=float, default=0.9)
parser.add_argument('--decoder_val_iters', type=int, default=None)
parser.add_argument('--decoder_grad_clip', type=float)
# wandb config
parser.add_argument('--no_wandb', dest='use_wandb', action='store_false')
# eval config
parser.add_argument('--eval_checkpoint', default=None)
args = parser.parse_args()
if args.project is None:
args.project = f'log_numbering'
args.identifier = f'{args.input_dim}classes_{args.dataset_size}samples_{args.model}_{args.hidden_dim}dim' + ('_DA' if args.rand_perm else '')
return args
class SetPredictionModel(pl.LightningModule):
def __init__(self, args):
super().__init__()
self.args = args
self.save_hyperparameters(args)
self.net = self.get_model(args.model)
self.trainset = data.ClassSpecificNumbering(n_samples=args.dataset_size, set_size=args.set_size, set_dim=args.input_dim, n_obj_per_sample=args.n_obj_per_sample, rand_perm=args.rand_perm)
self.valset = data.ClassSpecificNumbering(n_samples=6400, set_size=args.set_size, set_dim=args.input_dim, n_obj_per_sample=args.n_obj_per_sample)
self.testset = data.ClassSpecificNumbering(n_samples=64000, set_size=args.set_size, set_dim=args.input_dim, n_obj_per_sample=args.n_obj_per_sample)
def get_model(self, model_type):
hp = self.hparams
if 'dspn' in model_type:
input_enc_kwargs = dict(d_in=hp.input_dim, d_hid=hp.hidden_dim, d_latent=hp.latent_dim, set_size=hp.set_size, pool='fs')
inner_obj_kwargs = dict(d_in=hp.input_dim+hp.set_dim, d_hid=hp.hidden_dim, d_latent=hp.latent_dim, set_size=hp.set_size,
pool='fs', objective_type='mse_cat_input')
dspn_kwargs = dict(learn_init_set=False, set_dim=hp.set_dim, set_size=hp.set_size, momentum=hp.decoder_momentum, lr=hp.decoder_lr,
iters=hp.decoder_iters, grad_clip=hp.decoder_grad_clip, projection='simplex', implicit=model_type=='idspn')
net = models.DSPNBaseModel(input_enc_kwargs=input_enc_kwargs, inner_obj_kwargs=inner_obj_kwargs, dspn_kwargs=dspn_kwargs)
elif model_type == 'deepsets':
net = models.DSModel(hp.input_dim, hp.hidden_dim, hp.set_dim)
elif model_type == 'lstm':
net = models.LSTMModel(hp.input_dim, hp.hidden_dim, hp.set_dim)
elif model_type == 'transformer_with_pe':
net = models.TransformerWithPEModel(hp.input_dim, hp.hidden_dim, hp.set_dim, hp.set_size)
elif model_type == 'transformer_no_pe':
net = models.TransformerNoPEModel(hp.input_dim, hp.hidden_dim, hp.set_dim, hp.set_size)
elif model_type == 'transformer_rnd_pe':
net = models.TransformerRandomPEModel(hp.input_dim, hp.hidden_dim, hp.set_dim, hp.set_size)
return net
def forward(self, x):
input, gt_output = x
output = self.net(input)
if isinstance(output, tuple):
output, set_grad = output
else:
set_grad = None
return output, gt_output, set_grad
def training_step(self, batch, batch_idx):
return self.step(batch, batch_idx, '/train')
def validation_step(self, batch, batch_idx):
return self.step(batch, batch_idx, '/val')
def test_step(self, batch, batch_idx):
return self.step(batch, batch_idx, '/test')
def step(self, batch, batch_idx, suffix):
output, gt_output, set_grad = self(batch)
loss, indices = losses.hungarian_loss_numbering(
batch[0], output, gt_output,
num_workers=self.args.num_ray_workers,
ret_indices=True,
loss_type=self.args.loss.split('_')[-1])
loss = loss.mean(0)
micro_acc = losses.hungarian_micro_accuracy(output, gt_output, indices)
macro_acc = losses.hungarian_macro_accuracy(output, gt_output, indices)
log_dict = dict(loss=loss, micro_acc=micro_acc.mean(0), macro_acc=macro_acc.mean(0))
if set_grad is not None:
log_dict['grad_norm'] = set_grad.norm(dim=[1, 2]).mean()
self.log_dict({k+suffix: v for k,v in log_dict.items()})
return loss
def configure_optimizers(self):
opt = torch.optim.Adam(self.parameters(), lr=self.args.lr)
return opt
def train_dataloader(self):
return DataLoader(
self.trainset,
batch_size=self.args.batch_size,
shuffle=True,
num_workers=self.args.num_data_workers,
)
def val_dataloader(self):
return DataLoader(
self.valset,
batch_size=self.args.batch_size,
shuffle=False,
num_workers=self.args.num_data_workers,
)
def test_dataloader(self):
return DataLoader(
self.testset,
batch_size=self.args.batch_size,
shuffle=False,
num_workers=self.args.num_data_workers,
)
def train(args):
model = SetPredictionModel(args)
if args.num_ray_workers > 0:
ray.init(num_cpus=args.num_ray_workers, include_dashboard=False)
if args.use_wandb:
run = wandb.init(
name=args.name,
project=args.project,
reinit=False,
# settings=wandb.Settings(start_method='fork'),
)
run.define_metric('macro_acc/val', summary='max')
run.define_metric('micro_acc/val', summary='max')
logger = WandbLogger(log_model=True)
logger.watch(model.net)
wandb.config.update(args)
checkpoint_path = Path(args.checkpoint_path) / args.project / args.name
trainer = pl.Trainer(
max_epochs=args.epochs,
gpus=1,
num_nodes=1,
logger=logger if args.use_wandb else None,
callbacks=[
ModelCheckpoint(dirpath=checkpoint_path, monitor='macro_acc/val', mode='max'),
],
check_val_every_n_epoch=args.check_val_every_n_epoch,
)
trainer.fit(model)
trainer.test() # test best model
return model
def test(args, model=None, trainer=None):
if model is None:
model = SetPredictionModel.load_from_checkpoint(checkpoint_path=args.eval_checkpoint, args=args)
if trainer is None:
trainer = pl.Trainer(gpus=1, num_nodes=1, logger=None)
trainer.limit_test_batches = 1.0
trainer.test(model)
def main():
args = cmdline_args()
pl.seed_everything(args.seed)
if args.eval_checkpoint is None:
train(args)
else:
test(args)
if __name__ == '__main__':
main()