TSCAN/batch_generater.py at main · bb67ao/TSCAN · 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
import torch
from utils import fix_length
from einops import rearrange


def cf_train_quadkey(batch, data_source, max_len, sampler, quadkey_processor, TIME_processor, loc2quadkey, num_neg):
    # src_seq, trg_seq, t_mat, g_mat = zip(*batch)
    src_seq, trg_seq = zip(*batch)
    # t_mat_ = torch.stack(t_mat)
    # g_mat_ = torch.stack(g_mat)
    src_user_, src_locs_, src_quadkeys_, src_timecode_, src_lat_, src_lng_, src_times_ = [
    ], [], [], [], [], [], []
    data_size = []
    for e in src_seq:
        u_, l_, q_, t_, lat_, lng_, tg_, _ = zip(*e)
        src_user_.append(torch.tensor(u_))
        src_lat_.append(torch.tensor(lat_))
        src_lng_.append(torch.tensor(lng_))
        data_size.append(len(u_))
        src_locs_.append(torch.tensor(l_))
        q_ = quadkey_processor.numericalize(list(q_))
        # tg_ = TIME_processor.numericalize(list(tg_))
        src_timecode_.append(torch.tensor(tg_))
        src_quadkeys_.append(q_)
        src_times_.append(torch.tensor(t_))
    src_user_ = fix_length(src_user_, 1, max_len, 'train src seq')

    src_lat_ = fix_length(src_lat_, 1, max_len, 'train src seq')
    src_lng_ = fix_length(src_lng_, 1, max_len, 'train src seq')
    src_locs_ = fix_length(src_locs_, 1, max_len, 'train src seq')
    src_quadkeys_ = fix_length(src_quadkeys_, 2, max_len, 'train src seq')
    src_timecode_ = fix_length(src_timecode_, 2, max_len, 'train src seq')
    src_times_ = fix_length(src_times_, 1, max_len, 'train src seq')

    trg_locs_ = []
    trg_quadkeys_ = []
    trg_time_grams_ = []
    trg_times_ = []
    for i, seq in enumerate(trg_seq):
        pos = torch.tensor([[e[1]] for e in seq])
        pos_time_grams = torch.tensor([[e[6]] for e in seq])
        trg_times =  torch.tensor([[e[3]] for e in seq])
        neg = sampler(seq, num_neg, user=seq[0][0])
        pos_neg_locs = torch.cat([pos, neg], dim=-1)
        # pos_time_grams = pos_time_grams.repeat(1, 1+num_neg, 1)

        trg_times_.append(trg_times)
        trg_locs_.append(pos_neg_locs)
        trg_time_grams_.append(pos_time_grams)
        pos_neg_quadkey = []
        for l in range(pos_neg_locs.size(0)):
            q_key = []
            for loc_idx in pos_neg_locs[l]:
                q_key.append(loc2quadkey[loc_idx])
            pos_neg_quadkey.append(quadkey_processor.numericalize(q_key))
        # print(pos_neg_quadkey[0].size())
        # exit(0)
        trg_quadkeys_.append(torch.stack(pos_neg_quadkey))
    trg_locs_ = fix_length(trg_locs_, n_axies=2,
                           max_len=max_len, dtype='train trg seq')
    # print(trg_quadkeys_[0].size())
    # exit(0)
    trg_times_ = fix_length(trg_times_, n_axies=2,
                           max_len=max_len, dtype='train trg seq')
    trg_times_ = rearrange(
        rearrange(trg_times_, 'b n k -> k n b').contiguous(), 'k n b -> b (k n)')

    trg_time_grams_ = fix_length(
        trg_time_grams_, n_axies=3, max_len=max_len, dtype='train trg seq')
    trg_time_grams_ = rearrange(rearrange(
        trg_time_grams_, 'b n k l -> k n b l').contiguous(), 'k n b l -> b (k n) l')

    trg_locs_ = rearrange(
        rearrange(trg_locs_, 'b n k -> k n b').contiguous(), 'k n b -> b (k n)')

    trg_quadkeys_ = fix_length(
        trg_quadkeys_, n_axies=3, max_len=max_len, dtype='train trg seq')
    trg_quadkeys_ = rearrange(rearrange(
        trg_quadkeys_, 'b n k l -> k n b l').contiguous(), 'k n b l -> b (k n) l')

    # print(trg_time_grams_.size())
    # exit(0)
    # print(src_locs_.size(),src_quadkeys_.size(),trg_locs_.size(),trg_quadkeys_.size())

    # return src_locs_, src_quadkeys_, src_times_, t_mat_, g_mat_, trg_locs_, trg_quadkeys_, data_size
    return src_user_, src_locs_, src_quadkeys_, src_times_, src_timecode_, src_lat_, src_lng_, trg_locs_, trg_quadkeys_,trg_times_, trg_time_grams_, data_size


def cf_eval_quadkey(batch, data_source, max_len, sampler, quadkey_processor, timestamp_processor, loc2quadkey, num_neg):
    # src_seq, trg_seq, t_mat, g_mat = zip(*batch)
    # t_mat_ = torch.stack(t_mat)
    # g_mat_ = torch.stack(g_mat)
    src_seq, trg_seq = zip(*batch)
    src_user_, src_locs_, src_quadkeys_, src_timecode_, src_lat_, src_lng_, src_times_ = [
    ], [], [], [], [], [], []
    # src_locs_, src_quadkeys_, src_times_ = [], [], []
    data_size = []
    for e in src_seq:
        u_, l_, q_, t_, lat_, lng_, tg_, _ = zip(*e)
        src_user_.append(torch.tensor(u_))
        data_size.append(len(u_))
        src_locs_.append(torch.tensor(l_))
        src_lat_.append(torch.tensor(lat_))
        src_lng_.append(torch.tensor(lng_))
        q_ = quadkey_processor.numericalize(list(q_))
        # tg_ = timestamp_processor.numericalize(list(tg_))
        src_quadkeys_.append(q_)
        src_timecode_.append(torch.tensor(tg_))
        src_times_.append(torch.tensor(t_))
    src_user_ = fix_length(src_user_, 1, max_len, 'eval src seq')
    src_lat_ = fix_length(src_lat_, 1, max_len, 'eval src seq')
    src_lng_ = fix_length(src_lng_, 1, max_len, 'eval src seq')
    src_locs_ = fix_length(src_locs_, 1, max_len, 'eval src seq')
    src_quadkeys_ = fix_length(src_quadkeys_, 2, max_len, 'eval src seq')
    src_timecode_ = fix_length(src_timecode_, 2, max_len, 'eval src seq')
    src_times_ = fix_length(src_times_, 1, max_len, 'eval src seq')

    trg_locs_ = []
    trg_quadkeys_ = []
    trg_time_grams_ = []
    trg_times_ = []
    for i, seq in enumerate(trg_seq):
        pos = torch.tensor([[e[1]] for e in seq])
        pos_time_grams = torch.tensor([[e[6]] for e in seq])
        trg_times =  torch.tensor([[e[3]] for e in seq])
        neg_sample_from = [src_seq[i][-1]]
        # print(pos,neg_sample_from)
        # exit(0)
        neg = sampler(neg_sample_from, num_neg, user=neg_sample_from[0][0])
        # pos_time_grams = pos_time_grams.repeat(1, 1+num_neg, 1)
        # neg = sampler(seq, num_neg, user=seq[0][0])
        pos_neg_locs = torch.cat([pos, neg], dim=-1)

        trg_times_.append(trg_times)
        trg_locs_.append(pos_neg_locs)
        trg_time_grams_.append(pos_time_grams)
        pos_neg_quadkey = []
        for l in range(pos_neg_locs.size(0)):
            q_key = []
            for loc_idx in pos_neg_locs[l]:
                q_key.append(loc2quadkey[loc_idx])
            pos_neg_quadkey.append(quadkey_processor.numericalize(q_key))
        trg_quadkeys_.append(torch.stack(pos_neg_quadkey))

    trg_locs_ = fix_length(trg_locs_, n_axies=2,
                           max_len=max_len, dtype='eval trg loc')

    trg_times_ = fix_length(trg_times_, n_axies=2,
                           max_len=max_len, dtype='eval trg loc')
    trg_times_ = rearrange(
        rearrange(trg_times_, 'b n k -> k n b').contiguous(), 'k n b -> b (k n)')

    trg_time_grams_ = fix_length(
        trg_time_grams_, n_axies=3, max_len=max_len, dtype='eval trg loc')
    trg_time_grams_ = rearrange(rearrange(
        trg_time_grams_, 'b n k l -> k n b l').contiguous(), 'k n b l -> b (k n) l')

    trg_locs_ = rearrange(
        rearrange(trg_locs_, 'b n k -> k n b').contiguous(), 'k n b -> b (k n)')
    trg_quadkeys_ = fix_length(
        trg_quadkeys_, n_axies=3, max_len=max_len, dtype='eval trg loc')
    trg_quadkeys_ = rearrange(rearrange(
        trg_quadkeys_, 'b n k l -> k n b l').contiguous(), 'k n b l -> b (k n) l')

    # print(trg_time_grams_.size())
    # print(trg_quadkeys_.size())
    # exit(0)

    return src_user_, src_locs_, src_quadkeys_, src_times_, src_timecode_, src_lat_, src_lng_, trg_locs_, trg_quadkeys_,trg_times_, trg_time_grams_, data_size