OmiGraph/engine.py at main · ICTMCG/OmiGraph · 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
129
130
131
132
133
134
135
136
137
138
139
140
141
142
import math
import os
import sys
from typing import Iterable
import torch
import json
import utils.misc as utils
import numpy as np
import torch.distributed as dist
from sklearn.metrics import recall_score, precision_score, f1_score, accuracy_score, roc_auc_score

def to_device(item, device):
    if isinstance(item, torch.Tensor):
        return item.to(device)
    elif isinstance(item, list):
        return [to_device(i, device) for i in item]
    elif isinstance(item, dict):
        return {k: to_device(v, device) for k,v in item.items()}
    else:
        raise NotImplementedError("Call if you use other containers! type: {}".format(type(item)))


def train_one_epoch(model: torch.nn.Module, data_loader: Iterable,
                    optimizer: torch.optim.Optimizer, device: torch.device,
                    epoch: int, lr_scheduler = None, max_norm: float = 0, args=None, criterion_weight_dict=None):

    model.train()
    metric_logger = utils.MetricLogger(delimiter="  ")
    metric_logger.add_meter('lr', utils.SmoothedValue(window_size=1, fmt='{value:.6f}'))
    header = 'Epoch: [{}]'.format(epoch)
    print_freq = args.print_freq

    for samples in metric_logger.log_every(data_loader, print_freq, header):
        samples = to_device(samples, device)
        outputs = model(args, samples)
        if args.distributed:
            model_without_ddp = model.module
        else:
            model_without_ddp = model
        loss_dict = model_without_ddp.get_criterion(outputs, samples["label"])
        weight_dict = criterion_weight_dict
        losses = sum(loss_dict[k] * weight_dict[k] for k in loss_dict.keys() if k in weight_dict)

        # reduce losses over all GPUs for logging purposes
        loss_dict_reduced = utils.reduce_dict(loss_dict)
        loss_dict_reduced_unscaled = {f'{k}_unscaled': v
                                      for k, v in loss_dict_reduced.items()}
        loss_dict_reduced_scaled = {k: v * weight_dict[k]
                                    for k, v in loss_dict_reduced.items() if k in weight_dict}
        losses_reduced_scaled = sum(loss_dict_reduced_scaled.values())

        loss_value = losses_reduced_scaled.item()

        if not math.isfinite(loss_value):
            print("Loss is {}, stopping training".format(loss_value))
            print(loss_dict_reduced)
            sys.exit(1)

        # original backward function
        optimizer.zero_grad()
        losses.backward()
        if max_norm > 0:
            torch.nn.utils.clip_grad_norm_(model.parameters(), max_norm)
        optimizer.step()
        # lr_scheduler.step()

        metric_logger.update(loss=loss_value, **loss_dict_reduced_scaled, **loss_dict_reduced_unscaled)
        metric_logger.update(lr=optimizer.param_groups[0]["lr"])

    # gather the stats from all processes
    metric_logger.synchronize_between_processes()
    print("Averaged stats:", metric_logger)


@torch.no_grad()
def gather_together(data):
    world_size = utils.get_world_size()
    if world_size < 2:
        return [data]
    dist.barrier()
    gather_data = [None for _ in range(world_size)]
    dist.all_gather_object(gather_data, data)
    return gather_data


@torch.no_grad()
def evaluate(model, data_loader, device, args=None):
    model.eval()

    metric_logger = utils.MetricLogger(delimiter="  ")
    header = 'Test:'
    print_freq = args.print_freq

    y_true, y_pred, news_ids = [], [], []
    for samples in metric_logger.log_every(data_loader, print_freq, header):
        samples = to_device(samples, device)
        labels = samples["label"]
        outputs = model(args, samples)
        news_id = samples["news_id"]

        if isinstance(outputs, tuple):
            outputs = outputs[0]
        y_pred.extend(outputs.softmax(dim=1)[:, 1].flatten().tolist())
        y_true.extend(labels.flatten().tolist())
        news_ids.extend(news_id.flatten().tolist())

    merge_y_true = []
    for data in gather_together(y_true):
        merge_y_true.extend(data)

    merge_y_pred = []
    for data in gather_together(y_pred):
        merge_y_pred.extend(data)

    y_true, y_pred = np.array(merge_y_true), np.array(merge_y_pred)
    if args.save_preds_logits:
        prediction_file = os.path.join(args.output_dir + f"/{args.custom_log_name}.txt")
        with open(prediction_file, 'w') as f:
            for nid, true, pred in zip(news_ids, y_true, y_pred):
                # pred = 1 if pred > 0.5 else 0
                f.write(f"{int(nid)},\t{true},\t{pred}\n")

    all_metrics = {}
    try:
        all_metrics['auc'] = roc_auc_score(y_true, y_pred, average='macro')
        all_metrics['spauc'] = roc_auc_score(y_true, y_pred, average='macro', max_fpr=0.1)
    except:
        all_metrics['auc'] = 0
        all_metrics['spauc'] = 0
    y_pred = y_pred > 0.5
    all_metrics['mac_f1'] = f1_score(y_true, y_pred, average='macro')
    all_metrics['f1_real'], all_metrics['f1_fake'] = f1_score(y_true, y_pred, average=None)
    all_metrics['recall'] = recall_score(y_true, y_pred, average='macro')
    all_metrics['recall_real'], all_metrics['recall_fake'] = recall_score(y_true, y_pred, average=None)
    all_metrics['precision'] = precision_score(y_true, y_pred, average='macro')
    all_metrics['precision_real'], all_metrics['precision_fake'] = precision_score(y_true, y_pred, average=None)
    all_metrics['acc'] = accuracy_score(y_true, y_pred)

    for k, v in all_metrics.items():
        print(f"{k}: {v}")

    return json.dumps(all_metrics), all_metrics, all_metrics['mac_f1']