RFM/Run_RFM_WoW.py at main · ChenchenJT/RFM · 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
from RFMWoWDataset import *
from RFM import *
from torch import optim
from trainers.DefaultTrainer import *
import torch.backends.cudnn as cudnn
import argparse
import os
from data.Utils_WoW import *


def train(args):
    data_path = 'dataset/wizard_of_wikipedia/'

    if torch.cuda.is_available():
        torch.distributed.init_process_group(backend='NCCL', init_method='env://')

    cudnn.enabled = True
    cudnn.benchmark = True
    cudnn.deterministic = True
    print(torch.__version__)
    print(torch.version.cuda)
    print(cudnn.version())

    init_seed(123456)

    batch_size = 32

    output_path = 'model/' + 'wizard_of_wikipedia/'

    vocab2id, id2vocab, id2freq = load_vocab(data_path + 'wow_input_output.vocab', t=args.min_vocab_freq)

    if not os.path.exists(data_path + 'glove.6B.300d.txt' + '.dat'):
        prepare_embeddings(data_path + 'glove.6B.300d.txt')
    emb_matrix = load_embeddings(data_path + 'glove.6B.300d.txt', id2vocab, args.embedding_size)

    samples, query, passage = load_default(args.dataset, data_path + args.dataset + '.answer',
                                           data_path + args.dataset + '.passage',
                                           data_path + args.dataset + '.pool',
                                           data_path + args.dataset + '.qrel',
                                           data_path + args.dataset + '.query')

    train_samples, dev_samples, test_seen_samples, test_unseen_samples = split_data(args.dataset,
                                                                                    data_path + args.dataset + '.split',
                                                                                    samples)
    print("The number of train_samples:", len(train_samples))

    train_dataset = RFMWoWDataset(vocab2id, args.mode, train_samples, query, passage, args.min_window_size,
                                  args.num_windows, args.knowledge_len, args.context_len)

    model = RFM(args.min_window_size, args.num_windows, args.embedding_size, args.knowledge_len, args.context_len,
                args.hidden_size, vocab2id, id2vocab, max_dec_len=70,
                beam_width=1, emb_matrix=emb_matrix)
    init_params(model, escape='embedding')

    model_optimizer = optim.Adam(model.parameters(), lr=0.0001)

    trainer = DefaultTrainer(model, args.local_rank)

    Total_params = 0
    Trainable_params = 0
    NonTrainable_params = 0

    for param in model.parameters():
        mulValue = np.prod(param.size())
        Total_params += mulValue  # total parameters
        if param.requires_grad:
            Trainable_params += mulValue  # trainable parameters
        else:
            NonTrainable_params += mulValue  # non-trainable parameters

    print(f'Total params: {Total_params}')
    print(f'Trainable params: {Trainable_params}')
    print(f'Non-trainable params: {NonTrainable_params}')

    # for i in range(10):
    #     trainer.train_epoch('ds_train', train_dataset, collate_fn, batch_size, i, model_optimizer)

    for i in range(30):
        if i == 0:
            train_embedding(model)
        trainer.train_epoch('fb_mle_mcc_ds_train', train_dataset, collate_fn, batch_size, i, model_optimizer)
        # multi_schedule.step()
        trainer.serialize(i, output_path=output_path)


def test(args):
    data_path = 'dataset/wizard_of_wikipedia/'

    cudnn.enabled = True
    cudnn.benchmark = True
    cudnn.deterministic = True
    print(torch.__version__)
    print(torch.version.cuda)
    print(cudnn.version())

    init_seed(123456)

    batch_size = 32

    output_path = 'model/' + 'wizard_of_wikipedia/'

    vocab2id, id2vocab, id2freq = load_vocab(data_path + 'wow_input_output.vocab', t=args.min_vocab_freq)

    samples, query, passage = load_default(args.dataset, data_path + args.dataset + '.answer',
                                           data_path + args.dataset + '.passage',
                                           data_path + args.dataset + '.pool',
                                           data_path + args.dataset + '.qrel',
                                           data_path + args.dataset + '.query')

    train_samples, dev_samples, test_seen_samples, test_unseen_samples = split_data(args.dataset,
                                                                                    data_path + args.dataset + '.split',
                                                                                    samples)
    print("The number of test_seen_samples:", len(test_seen_samples))
    print("The number of test_unseen_samples:", len(test_unseen_samples))

    test_seen_dataset = RFMWoWDataset(vocab2id, args.mode, test_seen_samples, query, passage, args.min_window_size,
                                      args.num_windows, args.knowledge_len, args.context_len)

    test_unseen_dataset = RFMWoWDataset(vocab2id, args.mode, test_unseen_samples, query, passage, args.min_window_size,
                                        args.num_windows, args.knowledge_len, args.context_len)

    for i in range(30):
        print('epoch ' + str(i))
        file = output_path + 'model/' + str(i) + '.pkl'

        if os.path.exists(file):
            model = RFM(args.min_window_size, args.num_windows, args.embedding_size, args.knowledge_len,
                        args.context_len, args.hidden_size, vocab2id, id2vocab, max_dec_len=70, beam_width=1)
            model.load_state_dict(torch.load(file))
            trainer = DefaultTrainer(model, None)
            # trainer.test('test', dev_dataset, collate_fn, batch_size, i, output_path=output_path)
            # seen
            print('test_seen:')
            trainer.test('test', test_seen_dataset, collate_fn, batch_size, 100 + i, output_path=output_path,
                         test_type=args.test)
            # unseen
            print('test_unseen:')
            trainer.test('test', test_unseen_dataset, collate_fn, batch_size, 1000 + i, output_path=output_path,
                         test_type=args.test)


if __name__ == '__main__':
    parser = argparse.ArgumentParser()
    parser.add_argument("--local_rank", type=int, default=0)
    parser.add_argument("--mode", type=str, default='train')
    parser.add_argument("--test", type=str, default='SR')
    parser.add_argument("--dataset", type=str, default='wizard_of_wikipedia')
    parser.add_argument("--version", type=str, default='oracle')  # background version
    parser.add_argument("--embedding_size", type=int, default=300)  # embedding size
    parser.add_argument("--hidden_size", type=int, default=256)  # hidden size
    parser.add_argument("--min_window_size", type=int, default=4)  # the minimum size of slide window
    parser.add_argument("--num_windows", type=int, default=1)  # the stride of slide window
    parser.add_argument("--knowledge_len", type=int, default=256)  # background knowledge length
    parser.add_argument("--context_len", type=int, default=65)  # context length
    parser.add_argument("--min_vocab_freq", type=int, default=10)  # the minimum size of word frequency
    args = parser.parse_args()

    if args.mode == 'test':
        test(args)
    elif args.mode == 'train':
        train(args)