-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathtest.py
161 lines (142 loc) · 6.8 KB
/
test.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
import os
os.environ["CUDA_VISIBLE_DEVICES"] = "1"
import matplotlib.pyplot as plt
import numpy as np
from sklearn.manifold import TSNE
from sklearn.decomposition import PCA
import torch
from torch.autograd import Variable
from OpenKE.openke.config import Trainer, Tester
from OpenKE.openke.module.model import TransE
from OpenKE.openke.data import TestDataLoader, TrainDataLoader
from load_dataset import load_entity
from transemodel import embed_entity
def rank_sim(target, ent_embeds):
n_ent = ent_embeds.shape[0]
ent_sims = torch.zeros((n_ent, ))
batch_size = 16
begin = 0
while begin < n_ent:
end = min(n_ent, begin + batch_size)
ent_sims[begin: end] = torch.cosine_similarity(target, ent_embeds[begin: end], dim = -1)
begin += batch_size
sorted_idx = sorted(range(n_ent), key = lambda x: ent_sims[x], reverse = True)
return sorted_idx[1:11]
def rank_dist(target, ent_embeds):
n_ent = ent_embeds.shape[0]
ent_sims = torch.zeros((n_ent, ))
batch_size = 16
begin = 0
while begin < n_ent:
end = min(n_ent, begin + batch_size)
ent_sims[begin: end] = torch.cdist(target, ent_embeds[begin: end], p = 2)
begin += batch_size
sorted_idx = sorted(range(n_ent), key = lambda x: ent_sims[x])
return sorted_idx[:10]
# 将头实体、关系、尾实体的embedding降维画图
def plot_triplet(all_embeds, head_emb, rel_emb, tail_emb, title):
# data = torch.stack([head_emb, rel_emb, tail_emb], dim = 0)
data = torch.stack([head_emb, rel_emb], dim = 0)
data = torch.cat([data, tail_emb], dim = 0)
# tsne = TSNE(n_components=2, init='pca', random_state=0)
# data = tsne.fit_transform(data.detach().numpy())
# pca = PCA(n_components=2)
# pca.fit(all_embeds.detach().numpy())
# data = pca.transform(data.detach().numpy())
data = data[:, :2].detach().numpy()
# 画点和原点之间连接的直线
plt.figure(figsize = (7.2, 7.2))
plt.plot([0, data[0, 0]], [0, data[0, 1]], linestyle='--', label = 'head')
plt.plot([data[0, 0], data[0, 0] + data[1, 0]], [data[0, 1], data[0, 1] + data[1, 1]], linestyle='--', label = 'relation')
# plt.plot([0, data[i, 0]], [0, data[2, 1]], linestyle='--', label = 'tail')
for i in range(2, len(data)):
plt.plot([0, data[i, 0]], [0, data[i, 1]], linestyle='--', label = 'tail', c = 'green')
plt.legend(['head', 'relation', 'tail'])
# plt.title(title)
plt.savefig(f'result/test/{title}.png', dpi = 600)
if __name__ == '__main__':
print('Test TransE:')
# load data
data_path = 'data/'
model_name = './result/transe_12111305.ckpt'
# model_name = './result/transe_12101726.ckpt'
print('load model: ', model_name)
ent_dict, ent_list = load_entity(data_path + 'entity2id.txt')
rel_dict, rel_list = load_entity(data_path + 'relation2id.txt')
transe = TransE(
ent_tot = len(ent_list), # total number of entity
rel_tot = len(rel_list), # total number of relation
dim = 768,
p_norm = 1,
norm_flag = True
).cuda()
transe.load_checkpoint(model_name)
test_dataloader = TestDataLoader(data_path, "link", True)
# 获取transe训练得到的所有的embedding
ent_embeds = embed_entity(transe.ent_embeddings, len(ent_list))
rel_embeds = embed_entity(transe.rel_embeddings, len(rel_list))
# 获取测试样本
with open(data_path + 'test2id.txt', 'r') as f:
test_triplets = f.readlines()[1:]
# 测试集上的链接预测结果
# tester = Tester(model = transe, data_loader = test_dataloader, use_gpu = True)
# mrr, mr, hit10, hit3, hit1 = tester.run_link_prediction(type_constrain = True)
# print()
# 测试集中的前三个样本
# 打印模型预测的概率最高的前 10 个尾实体
# print('Tail entity prediction: ')
# for j in list(range(0, 1000, 50)):
# head, tail, relation = map(int, test_triplets[j].strip().split('\t'))
# print(ent_list[head], ' + ', rel_list[relation], ' = ', ent_list[tail])
# # print(ent_embeds.shape, rel_embeds.shape, )
# target = ent_embeds[head] + rel_embeds[relation]
# sorted_idx = rank_dist(target.unsqueeze(0), ent_embeds)
# for i, index in enumerate(sorted_idx[:10]):
# print(i, ': ', ent_list[index])
# print()
# # 使用 transe.predict 来预测
# print('transe.predict: ')
# for j, [_, data_tail] in enumerate(test_dataloader):
# if j == 101:
# break
# if j not in list(range(0, 1000, 50)): # 取三个测试样本
# continue
# score = transe.predict({
# 'batch_h': Variable(torch.from_numpy(data_tail['batch_h']).cuda()),
# 'batch_t': Variable(torch.from_numpy(data_tail['batch_t']).cuda()),
# 'batch_r': Variable(torch.from_numpy(data_tail['batch_r']).cuda()),
# 'mode': data_tail['mode']
# }) # score 是距离，距离越小越相似
# sorted_idx = sorted(range(len(ent_list)), key = lambda x: score[x])
# print('Head: ', ent_list[data_tail['batch_h'].item()], ', Relatioin: ', rel_list[data_tail['batch_r'].item()])
# for i, index in enumerate(sorted_idx[:10]):
# print(i, ': ', ent_list[index])
# print()
# 选取三个实体
# 打印和指定实体最相似的前10个实体
# print('Similar entity prediction: ')
# # ent_embeds = embed_entity(transe.ent_embeddings, len(ent_list)) # (15817, 200)
# for target in [ent_dict['花生'], ent_dict['价格'], ent_dict['好吃']] + list(range(10, 20)):
# print('Target: ', ent_list[target])
# sorted_idx = rank_sim(ent_embeds[target].unsqueeze(0), ent_embeds)
# for i, index in enumerate(sorted_idx):
# print(i, ': ', ent_list[index])
# print()
# 将头实体、关系、尾实体的embedding降维画图
# print('Plot triplet: ')
# head_emb = ent_embeds[ent_dict['花生']]
# for j in list(range(0, 1000, 50)):
all_embeds = torch.cat([ent_embeds, rel_embeds])
pca = PCA(n_components=2)
pca.fit(all_embeds.detach().numpy())
ent_embeds = pca.transform(ent_embeds.detach().numpy())
rel_embeds = pca.transform(rel_embeds.detach().numpy())
ent_embeds = torch.tensor(ent_embeds)
rel_embeds = torch.tensor(rel_embeds)
for j in list(range(1, 10, 2)):
head, tail, relation = map(int, test_triplets[j].strip().split('\t'))
target = ent_embeds[head] + rel_embeds[relation]
sorted_idx = rank_dist(target.unsqueeze(0), ent_embeds)[:10]
print(ent_list[head], ' + ', rel_list[relation], ' = ', ent_list[sorted_idx[0]])
title = '{}-{}-{}'.format(ent_list[head], rel_list[relation], ent_list[tail])
plot_triplet(torch.cat([ent_embeds, rel_embeds]), ent_embeds[head], rel_embeds[relation], ent_embeds[sorted_idx], title = title)