forked from zhouhanhanhan/EIGAT
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathutils.py
173 lines (135 loc) · 6.04 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
import torch
# from torchviz import make_dot, make_dot_from_trace
from models import SpKBGATModified, SpKBGATConvOnly
from layers import ConvKB
from torch.autograd import Variable
import torch.nn as nn
import numpy as np
import matplotlib.pyplot as plt
from matplotlib.lines import Line2D
from copy import deepcopy
from preprocess import build_data
from create_batch import Corpus
import random
import argparse
import os
import logging
import time
import pickle
CUDA = torch.cuda.is_available()
def save_model(model, name, epoch, folder_name, total_epoch):
torch.save(model.state_dict(),
(folder_name + "trained_present.pth"))
print("Saving Present Model")
if (epoch+1) == total_epoch:
print("Saving Model")
torch.save(model.state_dict(),
(folder_name + "trained_{}.pth").format(epoch))
print("Done saving Model")
else:
print("Model unsaved")
gat_loss_func = nn.MarginRankingLoss(margin=0.5)
def batch_gat_loss(gat_loss_func, train_indices, entity_embed, relation_embed, valid_invalid_ratio_gat):
len_pos_triples = int(
train_indices.shape[0] / (int(valid_invalid_ratio_gat) + 1))
pos_triples = train_indices[:len_pos_triples]
neg_triples = train_indices[len_pos_triples:]
pos_triples = pos_triples.repeat(int(valid_invalid_ratio_gat), 1)
source_embeds = entity_embed[pos_triples[:, 0]]
relation_embeds = relation_embed[pos_triples[:, 1]]
tail_embeds = entity_embed[pos_triples[:, 2]]
x = source_embeds + relation_embeds - tail_embeds
pos_norm = torch.norm(x, p=1, dim=1)
source_embeds = entity_embed[neg_triples[:, 0]]
relation_embeds = relation_embed[neg_triples[:, 1]]
tail_embeds = entity_embed[neg_triples[:, 2]]
x = source_embeds + relation_embeds - tail_embeds
neg_norm = torch.norm(x, p=1, dim=1)
if(CUDA):
y = -torch.ones(int(valid_invalid_ratio_gat) * len_pos_triples).cuda()
else:
y = -torch.ones(int(valid_invalid_ratio_gat) * len_pos_triples)
loss = gat_loss_func(pos_norm, neg_norm, y)
return loss
def GAT_Loss(train_indices, entity_embed, relation_embed, valid_invalid_ratio):
len_pos_triples = train_indices.shape[0] // (int(valid_invalid_ratio) + 1)
pos_triples = train_indices[:len_pos_triples]
neg_triples = train_indices[len_pos_triples:]
pos_triples = pos_triples.repeat(int(valid_invalid_ratio), 1)
source_embeds = entity_embed[pos_triples[:, 0]]
relation_embeds = relation_embed[pos_triples[:, 1]]
tail_embeds = entity_embed[pos_triples[:, 2]]
x = source_embeds + relation_embeds - tail_embeds
pos_norm = torch.norm(x, p=2, dim=1)
source_embeds = entity_embed[neg_triples[:, 0]]
relation_embeds = relation_embed[neg_triples[:, 1]]
tail_embeds = entity_embed[neg_triples[:, 2]]
x = source_embeds + relation_embeds - tail_embeds
neg_norm = torch.norm(x, p=2, dim=1)
y = torch.ones(int(valid_invalid_ratio)
* len_pos_triples).cuda()
loss = gat_loss_func(pos_norm, neg_norm, y)
return loss
def render_model_graph(model, Corpus_, train_indices, relation_adj, averaged_entity_vectors):
graph = make_dot(model(Corpus_.train_adj_matrix, train_indices, relation_adj, averaged_entity_vectors),
params=dict(model.named_parameters()))
graph.render()
def print_grads(model):
print(model.relation_embed.weight.grad)
print(model.relation_gat_1.attention_0.a.grad)
print(model.convKB.fc_layer.weight.grad)
for name, param in model.named_parameters():
print(name, param.grad)
def clip_gradients(model, gradient_clip_norm):
torch.nn.utils.clip_grad_norm_(model.parameters(), gradient_clip_norm)
for name, param in model.named_parameters():
if param.requires_grad:
print(name, "norm before clipping is -> ", param.grad.norm())
torch.nn.utils.clip_grad_norm_(param, gradient_clip_norm)
print(name, "norm beafterfore clipping is -> ", param.grad.norm())
def plot_grad_flow(named_parameters, parameters):
'''Plots the gradients flowing through different layers in the net during training.
Can be used for checking for possible gradient vanishing / exploding problems.
Usage: Plug this function in Trainer class after loss.backwards() as
"plot_grad_flow(self.model.named_parameters())" to visualize the gradient flow'''
ave_grads = []
max_grads = []
layers = []
for n, p in zip(named_parameters, parameters):
if(p.requires_grad) and ("bias" not in n):
layers.append(n)
ave_grads.append(p.grad.abs().mean())
max_grads.append(p.grad.abs().max())
plt.bar(np.arange(len(max_grads)), max_grads, alpha=0.1, lw=1, color="r")
plt.bar(np.arange(len(max_grads)), ave_grads, alpha=0.1, lw=1, color="b")
plt.hlines(0, 0, len(ave_grads) + 1, lw=2, color="g")
plt.xticks(range(0, len(ave_grads), 1), layers, rotation="vertical")
plt.xlim(left=0, right=len(ave_grads))
plt.ylim(bottom=-0.001, top=0.02) # zoom in on the lower gradient regions
plt.xlabel("Layers")
plt.ylabel("average gradient")
plt.title("Gradient flow")
plt.grid(True)
plt.legend([Line2D([0], [0], color="r", lw=4),
Line2D([0], [0], color="b", lw=4),
Line2D([0], [0], color="g", lw=4)], ['max-gradient', 'mean-gradient', 'zero-gradient'])
plt.savefig('initial.png')
plt.close()
def plot_grad_flow_low(named_parameters, parameters):
ave_grads = []
layers = []
for n, p in zip(named_parameters, parameters):
# print(n)
if(p.requires_grad) and ("bias" not in n):
layers.append(n)
ave_grads.append(p.grad.abs().mean())
plt.plot(ave_grads, alpha=0.3, color="b")
plt.hlines(0, 0, len(ave_grads) + 1, linewidth=1, color="k")
plt.xticks(range(0, len(ave_grads), 1), layers, rotation="vertical")
plt.xlim(xmin=0, xmax=len(ave_grads))
plt.xlabel("Layers")
plt.ylabel("average gradient")
plt.title("Gradient flow")
plt.grid(True)
plt.savefig('initial.png')
plt.close()