-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathrank_var_rate.py
129 lines (108 loc) · 4.99 KB
/
rank_var_rate.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
import numpy as np
import matplotlib.pyplot as plt
import pandas as pd
from pandas.plotting import table # EDIT: see deprecation warnings below
number_blocks = [6,12,24,16]
rank_importance = [[], [], [], [], []]
pre = 'rank_densenet121/densenet121_limit3'
total_importance = []
layer4_iomportance = []
compress_num = [[0], [0]*6, [0]*12, [0]*24, [0]*16]
compress_num2 = [[0], [0]*6, [0]*12, [0]*24, [0]*16]
total_data_num = [[0], [0]*6, [0]*12, [0]*24, [0]*16]
total_data_num2 = [[0], [0]*6, [0]*12, [0]*24, [0]*16]
compress_rate = [[0], [0]*6, [0]*12, [0]*24, [0]*16]
compress_rate2 = [[0], [0]*6, [0]*12, [0]*24, [0]*16]
y_max = 56
rank = pre + '/rank_conv%d'%(1) + '.npy'
data = np.load(rank)
mean_rank = data.mean()
var_rank = data.std()
sigma = 0.5
crit = mean_rank - sigma * var_rank
low_num = len(np.where(data <= crit)[0])
compress_num[0][0] = low_num
low_rate = low_num / len(data)
total_data_num[0][0] = len(data)
# importance = mean_rank / y_max
rank_importance[0].append([mean_rank, var_rank, low_num, low_rate])
# compress_rate[0].append(low_rate)
data_lenght = len(data)
total_importance
cnt=1
for i in range(4):
for j in range(number_blocks[i]):
cnt += 1
rank = pre + '/rank_conv%d'%(cnt) + '.npy'
data = np.load(rank)
mean_rank = data.mean()
var_rank = data.std()
crit = mean_rank - sigma * var_rank
low_num = len(np.where(data <= crit)[0])
total_data_num2[i+1][j] = len(data)
compress_num2[i+1][j] = low_num
low_rate = low_num / len(data)
# importance = mean_rank / y_max
rank_importance[i+1].append([mean_rank, var_rank, low_num, low_rate])
# compress_rate2[i+1].append(low_rate)
cnt += 1
rank = pre + '/rank_conv%d'%(cnt) + '.npy'
data = np.load(rank)
mean_rank = data.mean()
var_rank = data.std()
crit = mean_rank - sigma * var_rank
low_num = len(np.where(data <= crit)[0])
total_data_num[i+1][j] = len(data)
compress_num[i+1][j] = low_num
low_rate = low_num / len(data)
# importance = mean_rank / y_max
rank_importance[i+1].append([mean_rank, var_rank, low_num, low_rate])
# compress_rate[i+1].append(low_rate)
y_max = y_max // 2
if i != 3:
cnt += 1
rank = pre + '/rank_conv%d'%(cnt) + '.npy'
data = np.load(rank)
mean_rank = data.mean()
var_rank = data.std()
crit = mean_rank - sigma * var_rank
low_num = len(np.where(data <= crit)[0])
low_rate = low_num / len(data)
# importance = mean_rank / y_max
rank_importance[i+1].append([mean_rank, var_rank, low_num, low_rate])
# 2stage or 3stage or 4stage
comp_num = [[6], [3, 6, 2, 5, 6, 5], [3, 4, 3, 4, 5, 5, 4, 3, 7, 7, 6, 5], [5, 4, 5, 6, 5, 3, 4, 6, 5, 5, 5, 4, 6, 5, 5, 6, 4, 3, 5, 5, 4, 7, 6, 4], [5, 4, 6, 4, 5, 6, 6, 5, 4, 6, 4, 6, 6, 5, 7, 4]]
comp_num2 = [[0], [16, 26, 17, 23, 15, 19], [21, 21, 18, 19, 14, 16, 18, 20, 20, 16, 11, 19], [16, 18, 20, 17, 19, 23, 21, 17, 20, 19, 22, 19, 22, 21, 22, 20, 20, 19, 19, 18, 22, 20, 20, 18], [21, 20, 24, 18, 22, 20, 22, 27, 29, 27, 21, 21, 22, 23, 25, 21]]
comp_num3 = [[4], [2, 1, 1, 1, 1, 1], [2, 2, 2, 1, 2, 3, 2, 1, 2, 1, 2, 2], [1, 2, 2, 1, 2, 1, 0, 1, 1, 1, 1, 1, 2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], [1, 1, 1, 1, 1, 2, 1, 1, 1, 1, 1, 2, 1, 2, 0, 0]]
comp_num4 = [[0], [3, 3, 5, 7, 5, 7], [7, 4, 5, 7, 9, 6, 7, 6, 4, 6, 5, 9], [7, 4, 4, 8, 9, 4, 6, 5, 6, 5, 5, 6, 4, 7, 6, 5, 8, 8, 9, 7, 6, 6, 6, 3], [1, 2, 2, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 1, 0, 1]]
comp_num5 = [[3], [3, 1, 2, 2, 3, 4], [2, 3, 2, 2, 2, 2, 3, 2, 1, 1, 1, 2], [3, 2, 2, 1, 1, 2, 2, 2, 2, 2, 1, 1, 2, 1, 1, 1, 1, 2, 1, 2, 1, 1, 2, 1], [1, 2, 1, 2, 1, 2, 3, 1, 0, 1, 0, 1, 1, 0, 0, 0]]
comp_num6 = [[0], [7, 4, 10, 8, 13, 7], [7, 8, 9, 8, 9, 8, 9, 8, 8, 6, 5, 4], [7, 4, 8, 8, 8, 7, 9, 8, 9, 8, 6, 8, 7, 6, 8, 7, 8, 7, 9, 8, 5, 11, 9, 7], [1, 4, 3, 4, 0, 0, 1, 0, 0, 2, 2, 0, 1, 3, 0, 5]]
compress_num[0][0] += comp_num[0][0]
compress_num2[0][0] += comp_num2[0][0]
compress_num[0][0] += comp_num3[0][0]
compress_num2[0][0] += comp_num4[0][0]
compress_num[0][0] += comp_num5[0][0]
compress_num2[0][0] += comp_num6[0][0]
for i in range(4):
for j in range(number_blocks[i]):
compress_num[i+1][j] += comp_num[i+1][j]
compress_num2[i+1][j] += comp_num2[i+1][j]
for i in range(4):
for j in range(number_blocks[i]):
compress_num[i+1][j] += comp_num3[i+1][j]
compress_num2[i+1][j] += comp_num4[i+1][j]
for i in range(4):
for j in range(number_blocks[i]):
compress_num[i+1][j] += comp_num5[i+1][j]
compress_num2[i+1][j] += comp_num6[i+1][j]
compress_rate[0][0] = compress_num[0][0] / total_data_num[0][0]
for i in range(4):
for j in range(number_blocks[i]):
compress_rate[i+1][j] = compress_num[i+1][j] / total_data_num[i+1][j]
compress_rate2[i+1][j] = compress_num2[i+1][j] / total_data_num2[i+1][j]
print(compress_num)
print(compress_num2)
print(compress_rate)
print(compress_rate2)
# print(result)
# print(result2)