recbole-extension/custom_tests.py at master · alarca94/recbole-extension · 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
import time

import torch
from scipy.sparse import coo_matrix
import torch.nn.functional as F
import numpy as np
from numpy.lib.stride_tricks import as_strided
from numpy.linalg import eig

from scipy.sparse import csr_matrix, eye, hstack


def scipy_vs_torch_sparse():
    def unique_per_row(m):
        m_sorted = torch.sort(m)[0]
        duplicates = m_sorted[:, 1:] == m_sorted[:, :-1]
        m_sorted[:, 1:][duplicates] = 0
        return m_sorted

    def data_masks2(all_sessions):
        unique_values = unique_per_row(all_sessions)
        indices = unique_values.nonzero(as_tuple=True)
        col_ixs = unique_values[indices] - 1
        # crow_ixs = F.pad((unique_values > 0).sum(1), pad=(1, 0), mode="constant", value=0)
        return torch.sparse_coo_tensor(torch.stack([indices[0], col_ixs]), torch.ones_like(col_ixs), size=(len(all_sessions), 5))

    def data_masks(all_sessions):
        indptr, indices, data = [], [], []
        indptr.append(0)
        for j in range(len(all_sessions)):
            session = np.unique(all_sessions[j])
            length = len(session)
            s = indptr[-1]
            indptr.append((s + length))
            for i in range(length):
                indices.append(session[i] - 1)
                data.append(1)

        print(f'data: {data}')
        print(f'indices: {indices}')
        print(f'indptr: {indptr}')
        matrix = csr_matrix((data, indices, indptr), shape=(len(all_sessions), 5))
        print(matrix.todense())

    def data_masks_scipy(all_sessions):
        unique_values = unique_per_row(all_sessions)
        indices = unique_values.nonzero(as_tuple=True)
        col_ixs = unique_values[indices] - 1
        # crow_ixs neeeded for sparse_csr_tensor (not available in torch==1.8.1)
        crow_ixs = F.pad((unique_values > 0).sum(1), pad=(1, 0), mode="constant", value=0).cumsum(0)
        values = torch.ones_like(col_ixs).tolist()
        indices = col_ixs.tolist()
        indptr = crow_ixs.tolist()
        print(f'Values {values}')
        print(f'indices {indices}')
        print(f'indptr {indptr}')
        return csr_matrix((values, indices, indptr),
                          shape=(len(all_sessions), 5))

    s = [[4, 2, 1], [2, 4, 5, 1], [1, 3, 1]]
    data_masks(s)
    s2 = torch.tensor([[4, 2, 1, 0], [2, 4, 5, 1], [1, 3, 1, 0]])
    print(data_masks2(s2).to_dense())
    print('Hey')
    print(data_masks_scipy(s2).todense())


def test_hyperaugmentation():
    window_sizes = [1, 2]
    item_seq = torch.tensor([[4, 4, 78, 1, 3, 78, 0, 0, 0, 0], [4, 4, 78, 1, 3, 78, 0, 0, 0, 0]])
    node_mapping = {78: 0, 4: 1, 1: 2, 3: 3, 0: 4}

    seq = [[node_mapping[item] for item in item_seq[row].detach().numpy()] for row in range(item_seq.shape[0])]
    max_n_node = 8

    def regular_approach():
        c_seq = seq[0]
        start_time = time.time()
        h_A_in = [[], [], []]  # torch.zeros((max_n_node, n_edges))
        h_A_out = [[], [], []]  # torch.zeros((max_n_node, n_edges))
        it = 0
        for wsz in window_sizes:
            if len(c_seq) > wsz:
                for i in range(len(c_seq) - wsz):
                    ixs, vals = np.unique(c_seq[i:i + wsz], return_counts=True)
                    h_A_out[0].extend(ixs)
                    h_A_out[1].extend([it] * len(ixs))
                    h_A_out[2].extend(vals)
                    ixs, vals = np.unique(c_seq[i + wsz:i + wsz * 2], return_counts=True)
                    h_A_in[0].extend(ixs)
                    h_A_in[1].extend([it] * len(ixs))
                    h_A_in[2].extend(vals)
                    it += 1
            else:
                break

        h_A_out = coo_matrix((h_A_out[2], h_A_out[:2]), shape=(max_n_node, h_A_out[1][-1] + 1))
        h_A_in = coo_matrix((h_A_in[2], h_A_in[:2]), shape=(max_n_node, h_A_in[1][-1] + 1))
        print(f'Regular method {time.time() - start_time} s')
        invD_out = 1. / np.maximum(1., h_A_out.sum(1))
        invD_in = 1. / np.maximum(1., h_A_in.sum(1))
        invB_out = 1. / h_A_out.sum(0)
        invB_in = 1. / h_A_in.sum(0)

        u_A_out = h_A_in.multiply(invB_in).dot(h_A_out.multiply(invD_out).T)
        u_A_in = h_A_out.multiply(invB_out).dot(h_A_in.multiply(invD_in).T)
        u_A = np.concatenate([u_A_in.toarray(), u_A_out.toarray()]).T
        print(f'The input sequence is {c_seq}')
        # print(f'Subsets de salida: {list(zip(u_A_out[0], u_A_out[1], u_A_out[2]))}')
        # print(f'Subsets de entrada: {list(zip(u_A_in[0], u_A_in[1], u_A_in[2]))}')

        # u_A_out = coo_matrix((u_A_out[2], u_A_out[:2]), shape=(max_n_node, u_A_in[1][-1]+1))

        # print(u_A.sum(1))
        # print(f'A:\n{u_A}')
        # print(f'A:\n{h_A_out.dot(h_A_in.T).toarray().T}')
        # norm_h_out = h_A_out.multiply(invD_out).dot(h_A_out.multiply(invB_out).T)
        # norm_h_in = h_A_in.multiply(invD_in).dot(h_A_in.multiply(invB_in).T)
        # A = h_A_out.multiply(invD_out).dot(h_A_in.multiply(invB_in).T).toarray().T
        # print(eig(A)[0].max())
        # print(f'A:\n{A}')


    # TODO: Temporal decay (F-Hypergraph)
    # TODO: Backward attention (B-Hypergraph)

    def efficient_approach():
        start_time = time.time()
        def sliding_window(seq, wsz, sep):
            shape = seq.shape[:-1] + (seq.shape[-1] - wsz + 1, wsz)
            strides = seq.strides + (seq.strides[-1],)
            windows = as_strided(seq, shape=shape, strides=strides)
            cols = np.repeat(np.arange(shape[-2]-wsz), wsz)
            src_win = coo_matrix((np.repeat([1], len(cols)), (windows[:, :-wsz].ravel(), cols)), shape=(max_n_node, cols[-1]+1))
            trg_win = coo_matrix((np.repeat([1], len(cols)), (windows[:, (wsz+sep):].ravel(), cols)), shape=(max_n_node, cols[-1]+1))
            return src_win, trg_win
        # print(item_seq)
        eb = np.pad(seq, ((0, 0), [window_sizes[-1] - 1] * 2), constant_values=node_mapping[0])
        windows = list(zip(*[sliding_window(eb, wsz, 0) for wsz in window_sizes]))
        Ht = hstack(windows[0]).tocsr()
        Hh = hstack(windows[1]).tocsr()
        Ht[node_mapping[0], :] = 0
        Hh[node_mapping[0], :] = 0
        Ht[:, node_mapping[0]] = 0
        Hh[:, node_mapping[0]] = 0
        print(f'Efficient method {time.time() - start_time} s')
        print('hey')

    regular_approach()
    efficient_approach()

    print('Hey')


def test_oriented_graphs():
    s = [0, 1, 2, 0, 1, 0]
    h = np.zeros((len(np.unique(s)), len(s) - (2 - 1)))


if __name__ == '__main__':
    test_hyperaugmentation()