rescue/run_double.py at master · dbdmg/rescue · GitHub

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
import os
from pathlib import Path
import pickle
import wandb
from easydict import EasyDict as ed

import torch
from neural_net.cnn_configurations import TrainingConfig
from neural_net.utils import str2bool

import pytorch_lightning as pl
from pytorch_lightning.callbacks import ModelCheckpoint
from pytorch_lightning.callbacks.early_stopping import EarlyStopping
from pytorch_lightning.loggers import WandbLogger

from base_train import Satmodel, Double_Satmodel
import neural_net
from run_single import get_args


def train(args):
    hparams = TrainingConfig(**vars(args))
#                             , n_channels=24, mode='both')
#                             , only_burnt=False)

    if "SLURM_JOB_ID" in os.environ:
        print("Running in Slurm")
        hparams["job_id"] = os.environ["SLURM_JOB_ID"]
        hparams.num_workers = 4
        hparams.batch_size = 16

    hparams.model = {
        "name": "ConcatenatedModel",
        "model_dict": hparams.model
    }
    name = f'test_double-{args.model_name}_{args.key}'.lower()
    if args.losses: name += f"_{args.losses.lower()}"
    if args.seed is not None: name += f"_{args.seed}"

    outdir = Path(f"../data/new_ds_logs/Propaper/legion/{name}")#_imgnotnorm")

    outdir.mkdir(parents=True, exist_ok=True)

    if (outdir / 'bin_kpi.csv').exists() and (outdir / 'regr_kpi.csv').exists():
        print(f"Simulation already done ({name})")
        return

    if args.wandb_usr:
        run = wandb.init(reinit=True, project="rescue_paper", entity=args.wandb_usr, name=name, settings=wandb.Settings(start_method='fork'))

    print(f'Best checkpoints saved in "{outdir}"\n')

    hparams.checkpoint.dirpath = outdir

    if not torch.cuda.is_available():
        hparams.trainer.gpus = 0
        hparams.trainer.precision = 32

    if args.wandb_usr:
        logger = WandbLogger(save_dir=outdir, name=name)
        logger.log_hyperparams(hparams)
    else:
        logger = None


    #### 1st network ###################
    if not any(outdir.glob("bin*best*")):
        earlystopping_1 = EarlyStopping(**hparams.earlystopping)
        checkpoint_1 = ModelCheckpoint(**hparams.checkpoint, filename='binary_model-{epoch}')

        bin_model = Double_Satmodel(hparams, {'log_imgs': not args.discard_images, 'binary': True, 'log_res':not args.unlog_res})

        if args.wandb_usr:
            logger.watch(bin_model, log='all', log_freq=1)
        trainer = pl.Trainer(
            **hparams.trainer,
            max_epochs=hparams.epochs,
            logger=logger,
            callbacks=[checkpoint_1,
                       earlystopping_1
                       ],
        )
        trainer.fit(bin_model)
        best_path = Path(checkpoint_1.best_model_path)
        best = str(best_path.parent / f'{best_path.stem}_best{best_path.suffix}')
        best_path.rename(best)
    else:
        print("> Resuming from intermediate step.")
        best = str(next(outdir.glob("bin*best*")))

    #### 2nd network ###################
    if not any(outdir.glob("reg*best*")):
        earlystopping_2 = EarlyStopping(**hparams.earlystopping)
        checkpoint_2 = ModelCheckpoint(**hparams.checkpoint, filename='regression_model-{epoch}')

#         intermediate_chp = next(outdir.glob("reg*.ckpt"), None)
#         if intermediate_chp:
#             best = str(intermediate_chp)

        regr_model = Double_Satmodel.load_from_checkpoint(best,#checkpoint_1.best_model_path,
                                                          opt={'log_imgs': not args.discard_images, 'binary': False, 'log_res':not args.unlog_res}
                                                  )
        trainer = pl.Trainer(
            **hparams.trainer,
            max_epochs=hparams.epochs,
            logger=logger,
            callbacks=[checkpoint_2,
                       earlystopping_2
                       ],
        )
        trainer.fit(regr_model)

        trainer.test()
        best_path = Path(checkpoint_2.best_model_path)
        best_path.rename(best_path.parent / f'{best_path.stem}_best{best_path.suffix}')
        checkpoint_2.best_model_path = str(best_path)
    else:
        print("> Resuming weights for evaluation.")
        regr_model = Double_Satmodel.load_from_checkpoint(str(next(outdir.glob("reg*best*"))), hparams=hparams)
        trainer = pl.Trainer(**hparams.trainer, logger=logger).test(regr_model)

    if args.wandb_usr:
        wandb.finish()

if __name__ == '__main__':
    args = get_args()
    train(args)