-
Notifications
You must be signed in to change notification settings - Fork 2
/
datautil.py
186 lines (149 loc) · 6.69 KB
/
datautil.py
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
"""Helper functions for loading dataset"""
import scipy.sparse as sp
import numpy as np
from collections import defaultdict
from scipy.sparse import csr_matrix
import torch
def load_rating_file_to_list(filename):
"""Return **List** format user/group-item interactions"""
rating_list = []
lines = open(filename, 'r').readlines()
for line in lines:
contents = line.split()
# Each line: user item
rating_list.append([int(contents[0]), int(contents[1])])
return rating_list
def load_rating_file_to_matrix(filename, num_users=None, num_items=None):
"""Return **Matrix** format user/group-item interactions"""
if num_users is None:
num_users, num_items = 0, 0
lines = open(filename, 'r').readlines()
for line in lines:
contents = line.split()
u, i = int(contents[0]), int(contents[1])
num_users = max(num_users, u)
num_items = max(num_items, i)
mat = sp.dok_matrix((num_users + 1, num_items + 1), dtype=np.float32)
for line in lines:
contents = line.split()
if len(contents) > 2:
u, i, rating = int(contents[0]), int(contents[1]), int(contents[2])
if rating > 0:
mat[u, i] = 1.0
else:
u, i = int(contents[0]), int(contents[1])
mat[u, i] = 1.0
return mat
def load_negative_file(filename):
"""Return **List** format negative files"""
negative_list = []
lines = open(filename, 'r').readlines()
for line in lines:
negatives = line.split()[1:]
negatives = [int(neg_item) for neg_item in negatives]
negative_list.append(negatives)
return negative_list
def load_group_member_to_dict(user_in_group_path):
"""Return **Dict** format group-to-member-list mapping"""
group_member_dict = defaultdict(list)
lines = open(user_in_group_path, 'r').readlines()
for line in lines:
contents = line.split()
group = int(contents[0])
for member in contents[1].split(','):
group_member_dict[group].append(int(member))
return group_member_dict
def build_group_graph(group_data, num_groups):
"""Return group-level graph (**a weighted graph** with weights defined as ratio of common members and items)"""
matrix = np.zeros((num_groups, num_groups))
for i in range(num_groups):
group_a = set(group_data[i])
for j in range(i + 1, num_groups):
group_b = set(group_data[j])
overlap = group_a & group_b
union = group_a | group_b
# weight computation
matrix[i][j] = float(len(overlap) / len(union))
matrix[j][i] = matrix[i][j]
matrix = matrix + np.diag([1.0] * num_groups)
degree = np.sum(np.array(matrix), 1)
# \mathbf{D}^{-1} \dot \mathbf{A}
return np.dot(np.diag(1.0 / degree), matrix)
def build_hyper_graph(group_member_dict, group_train_path, num_users, num_items, num_groups, group_item_dict=None):
"""Return member-level hyper-graph"""
# Construct group-to-item-list mapping
if group_item_dict is None:
group_item_dict = defaultdict(list)
for line in open(group_train_path, 'r').readlines():
contents = line.split()
if len(contents) > 2:
group, item, rating = int(contents[0]), int(contents[1]), int(contents[2])
if rating > 0:
group_item_dict[group].append(item)
else:
group, item = int(contents[0]), int(contents[1])
group_item_dict[group].append(item)
def _prepare(group_dict, rows, axis=0):
nodes, groups = [], []
for group_id in range(num_groups):
groups.extend([group_id] * len(group_dict[group_id]))
nodes.extend(group_dict[group_id])
hyper_graph = csr_matrix((np.ones(len(nodes)), (nodes, groups)), shape=(rows, num_groups))
hyper_deg = np.array(hyper_graph.sum(axis=axis)).squeeze()
hyper_deg[hyper_deg == 0.] = 1
hyper_deg = sp.diags(1.0 / hyper_deg)
return hyper_graph, hyper_deg
# Two separate hypergraphs (user_hypergraph, item_hypergraph for hypergraph convolution computation)
user_hg, user_hg_deg = _prepare(group_member_dict, num_users)
item_hg, item_hg_deg = _prepare(group_item_dict, num_items)
for group_id, items in group_item_dict.items():
group_item_dict[group_id] = [item + num_users for item in items]
group_data = [group_member_dict[group_id] + group_item_dict[group_id] for group_id in range(num_groups)]
full_hg, hg_dg = _prepare(group_data, num_users + num_items, axis=1)
user_hyper_graph = torch.sparse.mm(convert_sp_mat_to_sp_tensor(user_hg_deg),
convert_sp_mat_to_sp_tensor(user_hg).t())
item_hyper_graph = torch.sparse.mm(convert_sp_mat_to_sp_tensor(item_hg_deg),
convert_sp_mat_to_sp_tensor(item_hg).t())
full_hyper_graph = torch.sparse.mm(convert_sp_mat_to_sp_tensor(hg_dg), convert_sp_mat_to_sp_tensor(full_hg))
print(
f"User hyper-graph {user_hyper_graph.shape}, Item hyper-graph {item_hyper_graph.shape}, Full hyper-graph {full_hyper_graph.shape}")
return user_hyper_graph, item_hyper_graph, full_hyper_graph, group_data
def convert_sp_mat_to_sp_tensor(x):
"""Convert `csr_matrix` into `torch.SparseTensor` format"""
coo = x.tocoo().astype(np.float32)
row = torch.Tensor(coo.row).long()
col = torch.Tensor(coo.col).long()
index = torch.stack([row, col])
data = torch.FloatTensor(coo.data)
return torch.sparse.FloatTensor(index, data, torch.Size(coo.shape))
def build_light_gcn_graph(group_item_net, num_groups, num_items):
"""Return item-level graph (**a group-item bipartite graph**)"""
adj_mat = sp.dok_matrix((num_groups + num_items, num_groups + num_items), dtype=np.float32)
adj_mat = adj_mat.tolil()
R = group_item_net.tolil()
adj_mat[:num_groups, num_groups:] = R
adj_mat[num_groups:, :num_groups] = R.T
adj_mat = adj_mat.todok()
# print(adj_mat, adj_mat.shape)
row_sum = np.array(adj_mat.sum(axis=1))
d_inv = np.power(row_sum, -0.5).flatten()
d_inv[np.isinf(d_inv)] = 0.
d_mat = sp.diags(d_inv)
# print(d_mat)
norm_adj = d_mat.dot(adj_mat)
norm_adj = norm_adj.dot(d_mat)
norm_adj = norm_adj.tocsr()
graph = convert_sp_mat_to_sp_tensor(norm_adj)
return graph.coalesce()
# Test code
# if __name__ == "__main__":
# g_m_d = {0: [0, 1, 2], 1: [2, 3], 2: [4, 5, 6]}
# g_i_d = {0: [0, 1], 1: [1, 2], 2: [3]}
# user_g, item_g, hg, g_data = build_hyper_graph(g_m_d, "", 7, 4, 3, g_i_d)
#
# print(user_g)
# print(item_g)
# print(hg)
# print()
# g = build_group_graph(g_data, 3)
# print(g)