PIAVAE/data.py at main · amazon-science/PIAVAE · 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
import os
import pandas as pd
from scipy import sparse
import numpy as np

import bottleneck as bn
import numpy as np


class DataLoader():
    '''
    Load Movielens-20m dataset
    '''
    def __init__(self, path):
        self.pro_dir = path
        assert os.path.exists(self.pro_dir), "Preprocessed files does not exist. Run data.py"

        self.n_items = self.load_n_items()

    def load_data(self, datatype='train'):
        if datatype == 'train':
            return self._load_train_data()
        elif datatype == 'validation':
            return self._load_tr_te_data(datatype)
        elif datatype == 'test':
            return self._load_tr_te_data(datatype)
        else:
            raise ValueError("datatype should be in [train, validation, test]")

    def load_n_items(self):
        unique_sid = list()
        with open(os.path.join(self.pro_dir, 'unique_sid.txt'), 'r') as f:
            for line in f:
                unique_sid.append(line.strip())
        n_items = len(unique_sid)
        return n_items

    def _load_train_data(self):
        path = os.path.join(self.pro_dir, 'train.csv')

        tp = pd.read_csv(path)
        n_users = tp['uid'].max() + 1

        rows, cols = tp['uid'], tp['sid']
        data = sparse.csr_matrix((np.ones_like(rows),
                                 (rows, cols)), dtype='float64',
                                 shape=(n_users, self.n_items))
        return data

    def _load_tr_te_data(self, datatype='test'):
        tr_path = os.path.join(self.pro_dir, '{}_tr.csv'.format(datatype))
        te_path = os.path.join(self.pro_dir, '{}_te.csv'.format(datatype))

        tp_tr = pd.read_csv(tr_path)
        tp_te = pd.read_csv(te_path)

        start_idx = min(tp_tr['uid'].min(), tp_te['uid'].min())
        end_idx = max(tp_tr['uid'].max(), tp_te['uid'].max())

        rows_tr, cols_tr = tp_tr['uid'] - start_idx, tp_tr['sid']
        rows_te, cols_te = tp_te['uid'] - start_idx, tp_te['sid']

        data_tr = sparse.csr_matrix((np.ones_like(rows_tr),
                                    (rows_tr, cols_tr)), dtype='float64', shape=(end_idx - start_idx + 1, self.n_items))
        data_te = sparse.csr_matrix((np.ones_like(rows_te),
                                    (rows_te, cols_te)), dtype='float64', shape=(end_idx - start_idx + 1, self.n_items))
        return data_tr, data_te

def get_count(tp, id):
    playcount_groupbyid = tp[[id]].groupby(id, as_index=False)
    count = playcount_groupbyid.size()
    return count

def filter_triplets(tp, min_uc=5, min_sc=0):
    if min_sc > 0:
        itemcount = get_count(tp, 'movieId')
        tp = tp[tp['movieId'].isin(itemcount.index[itemcount['size'] >= min_sc])]

    if min_uc > 0:
        usercount = get_count(tp, 'userId')
        tp = tp[tp['userId'].isin(usercount.index[usercount['size'] >= min_uc])]

    usercount, itemcount = get_count(tp, 'userId'), get_count(tp, 'movieId')
    return tp, usercount, itemcount

def split_train_test_proportion(data, test_prop=0.2):
    data_grouped_by_user = data.groupby('userId')
    tr_list, te_list = list(), list()

    np.random.seed(98765)

    for _, group in data_grouped_by_user:
        n_items_u = len(group)

        if n_items_u >= 5:
            idx = np.zeros(n_items_u, dtype='bool')
            idx[np.random.choice(n_items_u, size=int(test_prop * n_items_u), replace=False).astype('int64')] = True

            tr_list.append(group[np.logical_not(idx)])
            te_list.append(group[idx])

        else:
            tr_list.append(group)

    data_tr = pd.concat(tr_list)
    data_te = pd.concat(te_list)

    return data_tr, data_te

def numerize(tp, profile2id, show2id):
    uid = tp['userId'].apply(lambda x: profile2id[x])
    sid = tp['movieId'].apply(lambda x: show2id[x])
    return pd.DataFrame(data={'uid': uid, 'sid': sid}, columns=['uid', 'sid'])


# if __name__ == '__main__':

#     print("Load and Preprocess Movielens-20m dataset")
#     # Load Data
#     DATA_DIR = 'ml-20m/'
#     raw_data = pd.read_csv(os.path.join(DATA_DIR, 'ratings.csv'), header=0)
#     raw_data = raw_data[raw_data['rating'] > 3.5]

#     # Filter Data
#     raw_data, user_activity, item_popularity = filter_triplets(raw_data)

#     # Shuffle User Indices
#     unique_uid = user_activity.index
#     np.random.seed(98765)
#     idx_perm = np.random.permutation(unique_uid.size)
#     unique_uid = unique_uid[idx_perm]

#     n_users = unique_uid.size
#     n_heldout_users = 10000

#     # Split Train/Validation/Test User Indices
#     tr_users = unique_uid[:(n_users - n_heldout_users * 2)]
#     vd_users = unique_uid[(n_users - n_heldout_users * 2): (n_users - n_heldout_users)]
#     te_users = unique_uid[(n_users - n_heldout_users):]

#     train_plays = raw_data.loc[raw_data['userId'].isin(tr_users)]
#     unique_sid = pd.unique(train_plays['movieId'])

#     show2id = dict((sid, i) for (i, sid) in enumerate(unique_sid))
#     profile2id = dict((pid, i) for (i, pid) in enumerate(unique_uid))

#     pro_dir = os.path.join(DATA_DIR, 'pro_sg')

#     if not os.path.exists(pro_dir):
#         os.makedirs(pro_dir)

#     with open(os.path.join(pro_dir, 'unique_sid.txt'), 'w') as f:
#         for sid in unique_sid:
#             f.write('%s\n' % sid)

#     vad_plays = raw_data.loc[raw_data['userId'].isin(vd_users)]
#     vad_plays = vad_plays.loc[vad_plays['movieId'].isin(unique_sid)]

#     vad_plays_tr, vad_plays_te = split_train_test_proportion(vad_plays)

#     test_plays = raw_data.loc[raw_data['userId'].isin(te_users)]
#     test_plays = test_plays.loc[test_plays['movieId'].isin(unique_sid)]

#     test_plays_tr, test_plays_te = split_train_test_proportion(test_plays)

#     train_data = numerize(train_plays, profile2id, show2id)
#     train_data.to_csv(os.path.join(pro_dir, 'train.csv'), index=False)

#     vad_data_tr = numerize(vad_plays_tr, profile2id, show2id)
#     vad_data_tr.to_csv(os.path.join(pro_dir, 'validation_tr.csv'), index=False)

#     vad_data_te = numerize(vad_plays_te, profile2id, show2id)
#     vad_data_te.to_csv(os.path.join(pro_dir, 'validation_te.csv'), index=False)

#     test_data_tr = numerize(test_plays_tr, profile2id, show2id)
#     test_data_tr.to_csv(os.path.join(pro_dir, 'test_tr.csv'), index=False)

#     test_data_te = numerize(test_plays_te, profile2id, show2id)
#     test_data_te.to_csv(os.path.join(pro_dir, 'test_te.csv'), index=False)

#     print("Done!")