File size: 4,350 Bytes
98e2ea5
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import os
from os import path
import logging
from omegaconf import OmegaConf
import hydra
import hashlib
import json
import wandb
import torch

## Uncomment the following line to make the code deterministic and use CUBLAS_WORKSPACE_CONFIG=:4096:8
torch.use_deterministic_algorithms(True)
import random
import numpy as np

from experiment import Experiment

logging.basicConfig(format="%(asctime)s - %(message)s", level=logging.INFO)
logger = logging.getLogger()
os.environ["TOKENIZERS_PARALLELISM"] = "false"


def get_model_name(config):
    masked_copy = OmegaConf.masked_copy(
        config, ["datasets", "model", "trainer", "optimizer"]
    )
    encoded = json.dumps(OmegaConf.to_container(masked_copy), sort_keys=True).encode()
    # encoded['seed']=
    hash_obj = hashlib.md5()
    hash_obj.update(encoded)
    hash_obj.update(f"seed: {config.seed}".encode())

    model_hash = str(hash_obj.hexdigest())
    if len(config.datasets) > 1:
        dataset_name = "joint"
    else:
        dataset_name = list(config.datasets.keys())[0]
        if dataset_name == "litbank":
            cross_val_split = config.datasets[dataset_name].cross_val_split
            dataset_name += f"_cv_{cross_val_split}"

    key = f"_{config['key']}" if config["key"] != "" else ""
    model_name = f"{dataset_name}_{model_hash}{key}"

    return model_name


def main_train(config):
    if config.paths.model_name is None:
        model_name = get_model_name(config)
    else:
        model_name = config.paths.model_name

    config.paths.model_dir = path.join(
        config.paths.base_model_dir, config.paths.model_name_prefix + model_name
    )
    config.paths.best_model_dir = path.join(config.paths.model_dir, "best")

    for model_dir in [config.paths.model_dir, config.paths.best_model_dir]:
        if not path.exists(model_dir):
            os.makedirs(model_dir)

    if config.paths.model_path is None:
        config.paths.model_path = path.abspath(
            path.join(config.paths.model_dir, config.paths.model_filename)
        )
        config.paths.best_model_path = path.abspath(
            path.join(config.paths.best_model_dir, config.paths.model_filename)
        )

    if config.paths.best_model_path is None and (config.paths.model_path is not None):
        config.paths.best_model_path = config.paths.model_path

    # Dump config file
    config_file = path.join(config.paths.model_dir, "config.json")
    with open(config_file, "w") as f:
        f.write(json.dumps(OmegaConf.to_container(config), indent=4, sort_keys=True))

    return model_name


def main_eval(config):
    if config.paths.model_dir is None:
        raise ValueError

    best_model_dir = path.join(config.paths.model_dir, "best")
    if path.exists(best_model_dir):
        config.paths.best_model_dir = best_model_dir
    else:
        config.paths.best_model_dir = config.paths.model_dir

    config.paths.best_model_path = path.abspath(
        path.join(config.paths.best_model_dir, config.paths.model_filename)
    )


def set_seed(seed):
    random.seed(seed)
    torch.manual_seed(seed)
    torch.cuda.manual_seed_all(seed)

    np.random.seed(seed)
    os.environ["PYTHONHASHSEED"] = str(seed)

    torch.backends.cudnn.enabled = False
    torch.backends.cudnn.benchmark = False
    torch.backends.cudnn.deterministic = True


@hydra.main(config_path="conf", config_name="config")
def main(config):
    set_seed(config.seed)

    if config.train:
        model_name = main_train(config)
    else:
        main_eval(config)
        model_name = path.basename(path.normpath(config.paths.model_dir))
        # Strip prefix
        if model_name.startswith(config.paths.model_name_prefix):
            model_name = model_name[len(config.paths.model_name_prefix) :]

    if config.use_wandb:
        # Wandb Initialization
        try:
            wandb.init(
                id=model_name,
                project="Major Entity Tracking",
                config=dict(config),
                resume=True,
            )
        except:
            # Turn off wandb
            config.use_wandb = False

    logger.info(f"Model name: {model_name}")
    Experiment(config)


if __name__ == "__main__":
    import sys

    sys.argv.append(f"hydra.run.dir={path.dirname(path.realpath(__file__))}")
    sys.argv.append("hydra/job_logging=none")
    main()