knowledge-graph-question-answering/train.py at main · edwinthomas444/knowledge-graph-question-answering · 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
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
import sys
sys.path.append('./')

from dataset.wikidata import WikiDataDataset
from torch.utils.data import DataLoader
from model.model import RigelModel
from dataset.wikidata import Maps
import torch.nn as nn
from tqdm import tqdm
import torch
import json
import argparse
import torch.optim as optim
import torch.optim.lr_scheduler as lr_scheduler
from utils.scoring import multi_label_metrics
import warnings
warnings.simplefilter('ignore')

def config_parser(config_path):
    with open(config_path, 'r') as f:
        config = json.load(f)
    return config

def evaluate(model, dataloader, dataset):
    ep_preds, ep_gt = [], []

    for batch in tqdm(dataloader, total=len(dataloader)):
        # move the batch to gpu
        batch = tuple(t.to('cuda') for t in batch)
        inputs_er = {
            "span_embs":batch[4],
            "triplet_ids_tr":batch[0],
            "offsets_tr":batch[2],
            "attention_tr":batch[1],
            "qid_inds":batch[3],
            "qn_emb":batch[5]
        }
        with torch.no_grad():
            model.eval()
            out = model(**inputs_er)
            # remove copy of output tensors from computation graph
            ep_preds.append(out.detach().to('cpu'))
            ep_gt.append(batch[6].detach().to('cpu'))

    # compute scores
    result, avg_f1 = multi_label_metrics(
                                y_pred=torch.cat(ep_preds, dim=0).tolist(),
                                y_true=torch.cat(ep_gt, dim=0).tolist(),
                                thresh=0.5,
                                labels=dataset.entity_labels)

    return result, avg_f1


def main():
    '''
    example usage:
    python ./train.py --config './configs/base.json'
    '''
    parser = argparse.ArgumentParser()
    parser.add_argument('--config', type=str, required=True)
    args = parser.parse_args()
    configs = config_parser(args.config)


    # Define Dataset class
    # note: hidden emb dim must be same for both sentence and span detn models (i.e 768)
    train_dataset = WikiDataDataset(
                                    configs['dataset']['ent_file'],
                                    configs['dataset']['prop_file'],
                                    configs['dataset']['trip_file'],
                                    configs['dataset']['train_ds_file'],
                                    max_cand=configs['hparams']['max_cand'],
                                    max_spans=configs['hparams']['max_spans'],
                                    max_properties=configs['hparams']['max_properties'],
                                    span_detn_model=configs['hparams']['span_model'],
                                    sentence_emb_model=configs['hparams']['sentence_model'],
                                    emb_dim=configs['hparams']['emb_size'],
                                    split='train')

    val_dataset = WikiDataDataset(
                                    configs['dataset']['ent_file'],
                                    configs['dataset']['prop_file'],
                                    configs['dataset']['trip_file'],
                                    configs['dataset']['val_ds_file'],
                                    max_cand=configs['hparams']['max_cand'],
                                    max_spans=configs['hparams']['max_spans'],
                                    max_properties=configs['hparams']['max_properties'],
                                    span_detn_model=configs['hparams']['span_model'],
                                    sentence_emb_model=configs['hparams']['sentence_model'],
                                    emb_dim=configs['hparams']['emb_size'],
                                    split='val')

    test_dataset = WikiDataDataset(
                                    configs['dataset']['ent_file'],
                                    configs['dataset']['prop_file'],
                                    configs['dataset']['trip_file'],
                                    configs['dataset']['test_ds_file'],
                                    max_cand=configs['hparams']['max_cand'],
                                    max_spans=configs['hparams']['max_spans'],
                                    max_properties=configs['hparams']['max_properties'],
                                    span_detn_model=configs['hparams']['span_model'],
                                    sentence_emb_model=configs['hparams']['sentence_model'],
                                    emb_dim=configs['hparams']['emb_size'],
                                    split='test')
    # obtain the train, test and val dataloader
    train_dataloader = DataLoader(train_dataset, batch_size=configs['train']['batch_size'])
    val_dataloader = DataLoader(val_dataset, batch_size=configs['train']['batch_size'])
    test_dataloader = DataLoader(test_dataset, batch_size=configs['train']['batch_size'])

    model = RigelModel(
        triplet_size=train_dataset.unique_po,
        max_spans=configs['hparams']['max_spans'],
        max_cand=configs['hparams']['max_cand'],
        max_prop=configs['hparams']['max_properties'],
        num_entities=train_dataset.total_entities,
        max_hops=configs['hparams']['max_hops'],
        Ms=train_dataset.get_sparse_matrix(Maps.subj).to('cuda'),
        Mo=train_dataset.get_sparse_matrix(Maps.obj).to('cuda'),
        Mp=train_dataset.get_sparse_matrix(Maps.prop).to('cuda'),
        hdim=train_dataset.total_properties,
        emb_dim=configs['hparams']['emb_size']
    )
    print(model)
    model.to('cuda')

    # define loss
    loss = nn.BCELoss()

    # define total epochs
    total_epochs = configs['train']['epochs']

    # define lr schedulers and optimizers
    optimizer = optim.Adam(model.parameters(), lr=configs['train']['lr'])
    scheduler = lr_scheduler.ExponentialLR(optimizer, gamma=0.96)

    best_val, best_mpath = 0.0, ''
    # train for total number of epochs
    for epoch in range(0,total_epochs):
        ep_preds, ep_gt = [], []

        for batch in tqdm(train_dataloader, total=len(train_dataloader)):
            # move the batch to gpu
            batch = tuple(t.to('cuda') for t in batch)
            inputs_er = {
                "span_embs":batch[4],
                "triplet_ids_tr":batch[0],
                "offsets_tr":batch[2],
                "attention_tr":batch[1],
                "qid_inds":batch[3],
                "qn_emb":batch[5]
            }

            model.train()

            out = model(**inputs_er)
            out_loss = loss(out, batch[6])
            # add batch preds and gt

            # remove copy of output tensors from computation graph
            ep_preds.append(out.detach().to('cpu'))
            ep_gt.append(batch[6].detach().to('cpu'))

            optimizer.zero_grad()
            out_loss.backward()
            optimizer.step()

        # compute scores train
        result, train_f1 = multi_label_metrics(
                                    y_pred=torch.cat(ep_preds, dim=0).tolist(),
                                    y_true=torch.cat(ep_gt, dim=0).tolist(),
                                    thresh=0.5,
                                    labels=train_dataset.entity_labels)
        # save metrics for each epoch
        result.to_csv(f'./results/result1_train_{epoch}.csv')

        # obtain validation metrics
        val_results, val_f1 = evaluate(model, val_dataloader, val_dataset)
        val_results.to_csv(f'./results/result1_val_{epoch}.csv')

        if val_f1>=best_val:
            best_mpath = configs['train']['save_path'][:-3]+f'_best2.pt'
            # save checkpoint on best val scores
            torch.save({
                'epoch': epoch,
                'model_state_dict': model.state_dict(),
                'loss': out_loss.item(),
                }, best_mpath)

        print('Train f1: ', train_f1)
        print('Validation f1: ', val_f1)

        # step lr after each epoch
        scheduler.step()
        after_lr = optimizer.param_groups[0]["lr"]
        print(f'\nEpoch: {epoch}, Loss: {out_loss.item()}, LR: {after_lr}')

    # evaluate on the test set for best checkpoint (on eval)
    if best_mpath:
        checkpt = torch.load(best_mpath)
        model.load_state_dict(checkpt['model_state_dict'])
        test_results, test_f1 = evaluate(model, test_dataloader, test_dataset)
        test_results.to_csv(f'./results/result1_test.csv')
        print('\n Final Test f1: ', test_f1)

if __name__ == '__main__':
    main()