-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathevaluate.py
124 lines (101 loc) · 4.11 KB
/
evaluate.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
# -- coding:UTF-8
import numpy as np
import torch
import time
import pdb
import math
def largest_indices(ary, n):
"""Returns the n largest indices from a numpy array."""
flat = ary.flatten()
indices = np.argpartition(flat, -n)[-n:]
indices = indices[np.argsort(-flat[indices])]
return np.unravel_index(indices, ary.shape)
def hr_ndcg(indices_sort_top,index_end_i,top_k):
hr_topK=0
ndcg_topK=0
ndcg_max=[0]*top_k
temp_max_ndcg=0
for i_topK in range(top_k):
temp_max_ndcg+=1.0/math.log(i_topK+2)
ndcg_max[i_topK]=temp_max_ndcg
max_hr=top_k
max_ndcg=ndcg_max[top_k-1]
if index_end_i<top_k:
max_hr=(index_end_i)*1.0
max_ndcg=ndcg_max[index_end_i-1]
count=0
for item_id in indices_sort_top:
if item_id < index_end_i:
hr_topK+=1.0
ndcg_topK+=1.0/math.log(count+2)
count+=1
if count==top_k:
break
if max_hr == 0:
hr_t = 0
else:
hr_t=hr_topK/index_end_i
pre_t = hr_topK/top_k
ndcg_t=ndcg_topK/max_ndcg
# hr_t,ndcg_t,index_end_i,indices_sort_top
# pdb.set_trace()
return hr_t,pre_t,ndcg_t
def metrics_loss(model, test_val_loader_loss, batch_size):
start_time = time.time()
loss_sum=[]
loss_sum2=[]
for user, item_i, item_j in test_val_loader_loss:
user = user.cuda()
item_i = item_i.cuda()
item_j = item_j.cuda()
prediction_i, prediction_j,loss,loss2 = model(user, item_i, item_j)
loss_sum.append(loss.item())
loss_sum2.append(loss2.item())
# if np.isnan(loss2.item()).any():
# pdb.set_trace()
# pdb.set_trace()
elapsed_time = time.time() - start_time
test_val_loss1=round(np.mean(loss_sum),4)
test_val_loss=round(np.mean(loss_sum2),4)#round(np.mean(loss_sum[:-1]),4)#最后一个可能不满足一个batch，所以去掉这样loss就是一致的可以求mean了
str_print_val_loss=' val loss:'+str(test_val_loss)#+' time:'+str(round(elapsed_time,3))+' s'
# print(round(elapsed_time,3))
# print(test_val_loss1,test_val_loss)
return test_val_loss
def metrics(model, test_val_loader, top_k, num_negative_test_val, batch_size):
HR, NDCG = [], []
test_loss_sum=[]
# pdb.set_trace()
test_start_time = time.time()
for user, item_i, item_j in test_val_loader:
# start_time = time.time()
# pdb.set_trace()
user = user.cuda()
item_i = item_i.cuda()
item_j = item_j #index to split
prediction_i, prediction_j,loss_test,loss2_test = model(user, item_i, torch.cuda.LongTensor([0]))
test_loss_sum.append(loss2_test.item())
# pdb.set_trace()
elapsed_time = time.time() - test_start_time
print('time:'+str(round(elapsed_time,2)))
courrent_index=0
courrent_user_index=0
for len_i,len_j in item_j:
index_end_i=(len_i-len_j).item()
#pre_error=(prediction_i[0][courrent_index:(courrent_index+index_end_i)]- prediction_i[0][(courrent_index+index_end_i):(courrent_index+index_end_j)])#.sum()
#loss_test=nn.MSELoss((pre_error).sum())#-(prediction_i[0][courrent_index:(courrent_index+index_end_i)]- prediction_i[0][(courrent_index+index_end_i):(courrent_index+index_end_j)]).sigmoid().log()#.sum()
_, indices = torch.topk(prediction_i[0][courrent_index:(courrent_index+len_i)], top_k)
hr_t,ndcg_t=hr_ndcg(indices.tolist(),index_end_i,top_k)
# print(hr_t,ndcg_t,indices,index_end_i)
# pdb.set_trace()
HR.append(hr_t)
NDCG.append(ndcg_t)
courrent_index+=len_i
courrent_user_index+=1
test_loss=round(np.mean(test_loss_sum[:-1]),4)
return test_loss,round(np.mean(HR),4) , round(np.mean(NDCG),4)
# export ALL_PROXY=http://www-proxy.waseda.jp:8080/
# export http_proxy=$ALL_PROXY
# export https_proxy=$ALL_PROXY
# ./configure --prefix=${HOMW}/.local
# make
# make install