/
Data.py
196 lines (149 loc) · 6.89 KB
/
Data.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
187
188
189
190
191
192
193
194
195
196
import os
import networkx as nx
import numpy as np
import pickle as pkl
import scipy.sparse as sp
import sys
import torch
import codecs
TOPDIR = 'data/'
def add_self_loops(edge_list, size):
i = torch.arange(size, dtype=torch.int64).view(1, -1)
self_loops = torch.cat((i, i), dim=0)
edge_list = torch.cat((edge_list, self_loops), dim=1)
return edge_list
def get_degree(edge_list):
row, col = edge_list
deg = torch.bincount(row)
return deg
def normalize_adj(edge_list):
deg = get_degree(edge_list)
row, col = edge_list
deg_inv_sqrt = torch.pow(deg.to(torch.float), -0.5)
deg_inv_sqrt[deg_inv_sqrt == float('inf')] = 0.0
weight = torch.ones(edge_list.size(1))
v = deg_inv_sqrt[row] * weight * deg_inv_sqrt[col]
norm_adj = torch.sparse.FloatTensor(edge_list, v)
return norm_adj
def edgelist2normalized_adj(edge_list, size) -> (torch.Tensor, torch.sparse.FloatTensor):
edge_list = add_self_loops(edge_list, size)
norm_adj = normalize_adj(edge_list)
return edge_list, norm_adj
def edgelist2adj(edge_list) -> torch.sparse.FloatTensor:
v = torch.ones(edge_list.size(1))
adj = torch.sparse.FloatTensor(edge_list, v)
return adj
def index_to_mask(index, size):
mask = torch.zeros((size, ), dtype=torch.bool)
mask[index] = 1
return mask
def split_data(labels: torch.Tensor, n_train_per_class: int, n_val: int, seed) -> (torch.Tensor, torch.Tensor, torch.Tensor):
np.random.seed(seed)
n_class = int(torch.max(labels)) + 1
train_idx = np.array([], dtype=np.int64)
remains = np.array([], dtype=np.int64)
for c in range(n_class):
candidate = torch.nonzero(labels == c).T.numpy()[0]
np.random.shuffle(candidate)
train_idx = np.concatenate([train_idx, candidate[:n_train_per_class]])
remains = np.concatenate([remains, candidate[n_train_per_class:]])
np.random.shuffle(remains)
val_idx = remains[:n_val]
test_idx = remains[n_val:]
assert test_idx.shape[0] > val_idx.shape[0], ('No Test data', val_idx.shape[0], test_idx.shape[0])
train_mask = index_to_mask(train_idx, labels.size(0))
val_mask = index_to_mask(val_idx, labels.size(0))
test_mask = index_to_mask(test_idx, labels.size(0))
return train_mask, val_mask, test_mask
def preprocess_features(features:torch.Tensor):
rowsum = features.sum(dim=1, keepdim=True)
rowsum[rowsum == 0] = 1
features = features / rowsum
return features
class Data(object):
@staticmethod
def load(dataname):
top_dir = TOPDIR + dataname
labels = np.loadtxt(top_dir + '/labels.csv', dtype=np.int, delimiter=",")
features = np.loadtxt(top_dir + '/features.csv', dtype=np.float, delimiter=",")
edge_list = np.loadtxt(top_dir + '/edge_list.edg', dtype=np.int, delimiter=",").transpose()
split_setting = np.loadtxt(top_dir + '/split.txt', dtype=np.int, delimiter=",").tolist()
labels = torch.tensor(labels, dtype=torch.long)
features = torch.tensor(features, dtype=torch.float)
edge_list = torch.tensor(edge_list, dtype=torch.long)
# train_mask = torch.tensor(train_mask, dtype=torch.bool)
# valid_mask = torch.tensor(valid_mask, dtype=torch.bool)
# tests_mask = torch.tensor(tests_mask, dtype=torch.bool)
# print(train_mask.shape)
# use random mask
# train_mask, valid_mask, tests_mask = split_data(labels, 5, 50, seed)
data = Data(edge_list, features, labels, split_setting)
return data
def save(self, dataname):
top_dir = TOPDIR + dataname
if not os.path.exists(top_dir):
os.mkdir(top_dir)
np.savetxt(top_dir + '/labels.csv', self.labels.numpy(), '%d', delimiter=",")
np.savetxt(top_dir + '/features.csv', self.features.numpy(), '%f', delimiter=",")
np.savetxt(top_dir + '/edge_list.edg', self.raw_edge_list.numpy().transpose(), '%d', delimiter=",")
np.savetxt(top_dir + '/split.txt', self.split_setting, '%d', delimiter=",")
def __init__(self, edge_list: torch.Tensor, features: torch.Tensor, labels: torch.Tensor, split_setting: list):
self.raw_edge_list = edge_list
self.raw_adj = edgelist2adj(edge_list)
# normalized edge_list, normalized adj
self.norm_edge_list, self.norm_adj = edgelist2normalized_adj(edge_list, features.size(0))
self.features = features
self.labels = labels
self.split_setting = split_setting
self.num_features = features.size(1)
self.num_classes = int(torch.max(labels)) + 1
self.train_mask = None
self.valid_mask = None
self.tests_mask = None
self.update_mask()
self.num_train = torch.sum(self.train_mask.int(), dim=0).item()
self.num_valid = torch.sum(self.valid_mask.int(), dim=0).item()
self.num_tests = torch.sum(self.tests_mask.int(), dim=0).item()
# def to(self, device):
# self.adj = self.adj.to(device)
# self.edge_list = self.edge_list.to(device)
# self.features = self.features.to(device)
# self.labels = self.labels.to(device)
# self.train_mask = self.train_mask.to(device)
# self.valid_mask = self.valid_mask.to(device)
# self.tests_mask = self.tests_mask.to(device)
@property
def A(self):
return self.raw_adj
def update_mask(self, seed=None):
self.train_mask, self.valid_mask, self.tests_mask = split_data(self.labels, self.split_setting[0], self.split_setting[1], seed)
self.num_train = torch.sum(self.train_mask.int(), dim=0).item()
self.num_valid = torch.sum(self.valid_mask.int(), dim=0).item()
self.num_tests = torch.sum(self.tests_mask.int(), dim=0).item()
def print_statisitcs(self):
edge_list = self.raw_edge_list.numpy().transpose()
G = nx.Graph()
G.add_edges_from(edge_list)
print(" - statistics - ")
print("N:", len(nx.nodes(G)))
print("M:", len(nx.edges(G)))
assert len(nx.edges(G)) *2 == edge_list.shape[0]
print("M:", edge_list.shape[0], ' OK!')
print("number of components:", nx.number_connected_components(G))
print('features:', self.num_features)
print('classes:', self.num_classes)
# counts for each label
def label_counts(data: Data, mask=None):
counts = np.zeros(data.num_classes, dtype=np.int)
labels = data.labels.numpy()
for i in range(labels.shape[0]):
if mask is not None:
if not mask[i]:
continue
counts[int(labels[i])] += 1
return counts
print('data label balance:', label_counts(self))
print('num_train:', self.num_train, label_counts(self, self.train_mask))
print('num_valid:', self.num_valid, label_counts(self, self.valid_mask))
print('num_tests:', self.num_tests, label_counts(self, self.tests_mask))
print(" - ")