-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathvisualize_attention.py
110 lines (92 loc) · 3.89 KB
/
visualize_attention.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
import matplotlib
matplotlib.use('AGG')
from matplotlib import pyplot as plt
from matplotlib.font_manager import FontProperties
import os
font_file = 'DroidSansFallbackFull.ttf'
my_font = FontProperties(fname=font_file)
def visualize_attention_per_head(tmp_Vocab, tower_mems_id_np, attn_prob, next_word_index, file_path, length):
# todo 作图
head_num = 10
xLabel = tmp_Vocab.get_symbols([tower_mems_id_np[0][0][i][0] for i in range(-100, 0, 1)])
yLabel = list(range(0, 16))
datas = []
for l in range(head_num): #head
data = []
for i in range(len(yLabel)):
temp = []
for j in range(len(xLabel)):
k = attn_prob[0][i][0][-100+j][0][l]
temp.append(k)
data.append(temp)
datas.append(data)
# 作图阶段
fig = plt.figure(figsize=(48, 48))
for k in range(head_num):
ax = fig.add_subplot(5, 2, k+1)
# plt.subplots_adjust(left=(k%2)*0.5+0.05, right=(k%2)*0.5+0.45)
plt.subplots_adjust(left=0.05, right=0.95, top=0.95, bottom=0.05)
# 定义横纵坐标的刻度
ax.set_yticks(range(len(yLabel)))
ax.set_yticklabels(yLabel)
ax.set_xticks(range(len(xLabel)))
ax.set_xticklabels(xLabel, fontproperties=my_font, rotation=45, size=5)
# 作图并选择热图的颜色填充风格，这里选择hot
im = ax.imshow(datas[k], cmap=plt.cm.hot_r)
plt.title(tmp_Vocab.get_sym(next_word_index)+' head:{}'.format(k), fontproperties=my_font)
# 增加右侧的颜色刻度条
# plt.colorbar(im)
# 增加标题
# file_path = 'tmp_img'
if (os.path.exists(file_path) == False):
os.makedirs(file_path)
plt.savefig('{}/{}.png'.format(file_path, str(length)+tmp_Vocab.get_sym(next_word_index)))
# # # show
# plt.show()
def visualize_attention_per_layer(tmp_Vocab, tower_mems_id_np, attn_prob, next_word_index, file_path, length):
# todo 作图
head_num = 10
layer_num = 16
xLabel = tmp_Vocab.get_symbols([tower_mems_id_np[0][0][i][0] for i in range(100)])
yLabel = list(range(0, 10))
datas = []
for i in range(layer_num):
data = []
for l in range(head_num): # head
temp = []
for j in range(len(xLabel)):
k = attn_prob[0][i][0][-100+j][0][l]
temp.append(k)
data.append(temp)
datas.append(data)
# 作图阶段
fig = plt.figure(figsize=(48, 36))
for k in range(layer_num):
ax = fig.add_subplot(8, 2, k+1)
# plt.subplots_adjust(left=(k%2)*0.5+0.05, right=(k%2)*0.5+0.45)
plt.subplots_adjust(left=0.05, right=0.95, top=0.95, bottom=0.05)
# 定义横纵坐标的刻度
ax.set_yticks(range(len(yLabel)))
ax.set_yticklabels(yLabel)
ax.set_xticks(range(len(xLabel)))
ax.set_xticklabels(xLabel, fontproperties=my_font, rotation=45, size=5)
# 作图并选择热图的颜色填充风格，这里选择hot
im = ax.imshow(datas[k], cmap=plt.cm.hot_r)
plt.title(tmp_Vocab.get_sym(next_word_index)+' layer:{}'.format(k), fontproperties=my_font)
if (os.path.exists(file_path) == False):
os.makedirs(file_path)
plt.savefig('{}/{}.png'.format(file_path, str(length)+tmp_Vocab.get_sym(next_word_index)))
# # # show
# plt.show()
def visualize_prob(tmp_Vocab, tmp_list, file_path, length):
plt.figure()
index_list = sorted(range(len(tmp_list)), key=lambda k: tmp_list[k], reverse=True)[:10]
xlist = tmp_Vocab.get_symbols(index_list)
ylist = []
for i in range(len(index_list)):
ylist.append(tmp_list[index_list[i]])
plt.bar(range(len(xlist)), ylist)
plt.xticks(range(len(xlist)), xlist, fontproperties=my_font, rotation=0, size=5)
if (os.path.exists(file_path) == False):
os.makedirs(file_path)
plt.savefig('{}.png'.format(os.path.join(file_path, str(length))))