-
Notifications
You must be signed in to change notification settings - Fork 1
/
utils.py
178 lines (134 loc) · 6.55 KB
/
utils.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
# +
import matplotlib.pyplot as plt
import seaborn as sns
import torch
import torch.nn.functional as F
from torch_geometric.utils import degree, add_remaining_self_loops, k_hop_subgraph, to_networkx, subgraph
from torch_scatter import scatter
import torch_sparse
from torch_sparse import coalesce
import networkx as nx
def get_id(n, edge_index):
index = torch.tensor([list(range(n)), list(range(n))], device=edge_index.device).long()
value = torch.ones_like(index[0]).float()
return coalesce(index, value, m=n, n=n)
def power(edge_index, edge_weight, n, k):
if k == 0:
pow_edge_index, pow_edge_weight = get_id(n, edge_index)
return pow_edge_index, pow_edge_weight
pow_edge_index = edge_index.clone()
pow_edge_weight = edge_weight.clone()
for _ in range(k - 1):
pow_edge_index, pow_edge_weight = torch_sparse.spspmm(edge_index, edge_weight, pow_edge_index, pow_edge_weight, n, n, n)
return pow_edge_index, pow_edge_weight
def power_x(edge_index, edge_weight, x, n, k):
for _ in range(k):
x = torch_sparse.spmm(edge_index, edge_weight, n, n, x)
return x
def gcn_norm(edge_index, edge_weight, n, return_inv_sqrt=False):
deg = degree(edge_index[0], num_nodes=n)
deg_inv_sqrt = deg.pow(-0.5)
deg_inv_sqrt.masked_fill_(deg_inv_sqrt == float('inf'), 0)
if return_inv_sqrt:
return deg_inv_sqrt[edge_index[0]] * edge_weight * deg_inv_sqrt[edge_index[1]], deg_inv_sqrt
return deg_inv_sqrt[edge_index[0]] * edge_weight * deg_inv_sqrt[edge_index[1]]
def sage_norm(edge_index, edge_weight, n):
deg = degree(edge_index[0], num_nodes=n)
deg_inv = deg.pow(-1.0)
deg_inv.masked_fill_(deg_inv == float('inf'), 0)
return edge_weight * deg_inv[edge_index[1]]
# +
@torch.no_grad()
def test_err_by_deg(model, data):
model.eval()
pred = model(data.x, data.edge_index, data.edge_attr).argmax(dim=-1)
edge_index, _ = add_remaining_self_loops(data.edge_index)
edge_weight = torch.ones_like(edge_index[0]).float()
edge_weight = gcn_norm(edge_index, edge_weight, data.x.size(0))
deg = torch.zeros_like(data.y).float()
deg = torch.scatter_add(deg, 0, edge_index[0], edge_weight)
errs = []
degs = []
for mask in [data.train_mask, data.val_mask, data.test_mask]:
mismatches = (pred[mask] != data.y[mask]).float()
errs.append(mismatches)
degs.append(deg[mask])
return errs, degs
@torch.no_grad()
def test_loss_by_deg(model, data):
model.eval()
out = model(data.x, data.edge_index, data.edge_attr)
edge_index, _ = add_remaining_self_loops(data.edge_index)
edge_weight = torch.ones_like(edge_index[0]).float()
edge_weight = gcn_norm(edge_index, edge_weight, data.x.size(0))
deg = torch.zeros_like(data.y).float()
deg = torch.scatter_add(deg, 0, edge_index[0], edge_weight)
errs = []
degs = []
for mask in [data.train_mask, data.val_mask, data.test_mask]:
pred = F.log_softmax(out[mask], dim=1)
target = data.y[mask]
err = -pred[range(target.size(0)), target]
errs.append(err)
degs.append(deg[mask])
return errs, degs
# -
def label_homo(data, k):
edge_index, _ = add_remaining_self_loops(data.edge_index)
edge_weight = torch.ones_like(edge_index[0]).float()
edge_weight = gcn_norm(edge_index, edge_weight, data.x.size(0))
deg = torch.zeros_like(data.y).float()
deg = torch.scatter_add(deg, 0, edge_index[0], edge_weight)
homo = torch.zeros_like(data.y).float()
pow_edge_index, pow_edge_weight = power(edge_index, edge_weight, data.x.size(0), k)
c_mask = data.y[pow_edge_index[0]] == data.y[pow_edge_index[1]]
homo = torch.scatter_add(homo, 0, pow_edge_index[0, c_mask], pow_edge_weight[c_mask]) / deg
return [homo[mask] for mask in [data.train_mask, data.val_mask, data.test_mask]]
def feature_homo(data, k):
edge_index, _ = add_remaining_self_loops(data.edge_index)
edge_weight = torch.ones_like(edge_index[0]).float()
edge_weight = gcn_norm(edge_index, edge_weight, data.x.size(0))
deg = torch.zeros_like(data.y).float()
deg = torch.scatter_add(deg, 0, edge_index[0], edge_weight)
homo = torch.zeros_like(data.y).float()
pow_edge_index, pow_edge_weight = power(edge_index, edge_weight, data.x.size(0), k)
x_lowrank = torch.svd(data.x)[0][:, :2]
dists = torch.norm(x_lowrank[pow_edge_index[0]] - x_lowrank[pow_edge_index[1]], dim=1)
homo = torch.scatter_add(homo, 0, pow_edge_index[0], dists * pow_edge_weight)
return [homo[mask] for mask in [data.train_mask, data.val_mask, data.test_mask]]
def influence_scores(data, k):
edge_index, _ = add_remaining_self_loops(data.edge_index)
edge_weight = torch.ones_like(edge_index[0]).float()
edge_weight = gcn_norm(edge_index, edge_weight, data.x.size(0))
edge_index, edge_weight = power(edge_index, edge_weight, data.x.size(0), k)
deg = torch.zeros_like(data.y).float()
deg = torch.scatter_add(deg, 0, edge_index[0], edge_weight)
labeled_idx = data.train_mask.float().reshape(-1, 1)
labeled_influence = torch_sparse.spmm(edge_index, edge_weight, data.x.size(0), data.x.size(0), labeled_idx).flatten()
self_influence = edge_weight[edge_index[0] == edge_index[1]]
return [labeled_influence[mask] / deg[mask] for mask in [data.train_mask, data.val_mask, data.test_mask]]
def get_compatibility_matrix(y, edge_index, edge_weight=None):
"""
Return the weighted compatibility matrix, according to the weights in the provided adjacency matrix.
"""
src, dst = edge_index
num_classes = torch.unique(y).shape[0]
H = torch.zeros((num_classes, num_classes))
for i in range(num_classes):
for j in range(num_classes):
mask = (y == i)[src] & (y == j)[dst]
H[i, j] = edge_weight[mask].sum()
return torch.nn.functional.normalize(H, p=1)
def scatter(x, high_degs_mask, low_degs_mask, data):
low_deg_spread = 0
high_deg_spread = 0
for c in range(data.y.max() + 1):
mask = low_degs_mask & (data.y[data.test_mask] == c)
if mask.sum() != 0:
centered_x = x[mask] - x[mask].mean(dim=0).reshape(1, -1)
low_deg_spread += centered_x.t() @ centered_x / low_degs_mask.sum()
mask = high_degs_mask & (data.y[data.test_mask] == c)
if mask.sum() != 0:
centered_x = x[mask] - x[mask].mean(dim=0).reshape(1, -1)
high_deg_spread += centered_x.t() @ centered_x / high_degs_mask.sum()
return torch.trace(low_deg_spread), torch.trace(high_deg_spread)