-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathevaluator.py
176 lines (152 loc) · 7.02 KB
/
evaluator.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
174
import sys
import copy
import random
import time
import numpy as np
from tqdm import tqdm
import sampler
def predict_top_k_with_item_set(item_set, model, sess, us, seqs, batch_size, user_train=None, user_valid=None,
user_test=None, convert_hist_item_id=False, async_submitter=None):
items = np.array(sorted(list(item_set)))
if user_train:
top_k_values, top_k_indices = model.predict_top_k(sess, us, seqs, items, batch_size=batch_size,
user_hist_train=user_train, user_hist_val=user_valid,
user_hist_test=user_test,
convert_hist_item_id=convert_hist_item_id,
async_submitter=async_submitter)
else:
top_k_values, top_k_indices = model.predict_top_k(sess, us, seqs, items, batch_size=batch_size)
if async_submitter:
return None, None
# Convert back
top_k_indices_shape = np.shape(top_k_indices)
top_k_indices = top_k_indices.flatten()
top_k_indices = items[top_k_indices]
top_k_indices = np.reshape(top_k_indices, top_k_indices_shape)
return top_k_values, top_k_indices
def build_seq(train, valid, test, u, target_timestamp, args):
seq = np.zeros([args.maxlen], dtype=np.int32)
timestamp = np.zeros([args.maxlen], dtype=np.int64)
idx = args.maxlen - 1
if test is not None:
if args.load_timestamp:
seq[idx] = test[u][0][0]
timestamp[idx] = test[u][0][1]
else:
seq[idx] = test[u][0]
idx -= 1
if valid is not None:
if args.load_timestamp:
seq[idx] = valid[u][0][0]
timestamp[idx] = valid[u][0][1]
else:
seq[idx] = valid[u][0]
idx -= 1
for i in reversed(train[u]):
if args.load_timestamp:
seq[idx] = i[0]
timestamp[idx] = i[1]
else:
seq[idx] = i
idx -= 1
if idx == -1: break
if args.load_timestamp:
if args.compute_hstu_time_interval:
interval = sampler.compute_hstu_time_interval(timestamp, target_timestamp, args.hstu_time_interval_divisor,
max_value=args.time_interval_attention_max_interval)
else:
interval = sampler.compute_interval(timestamp, max_value=args.time_interval_attention_max_interval)
else:
interval = None
return seq, interval
def exact_evaluate(model, dataset, args, sess, mode='test', batch_size=0, sample_user_num=10000,
evaluate_user=(), evaluate_item=(), eval_item_not_in_history=False):
train, valid, test, user_num, item_num = dataset
NDCG = 0.0
valid_user = 0.0
HT = 0.0
users = list(evaluate_user) if len(evaluate_user) > 0 else range(1, user_num + 1)
if sample_user_num > 0 and len(users) > sample_user_num:
users = random.sample(users, sample_user_num)
eval_user_num = sample_user_num
else:
eval_user_num = len(users)
us = []
seqs = []
intervals = []
target_item_ids = []
print('Preparing evaluate data...')
for u in tqdm(users, total=eval_user_num, leave=False, ncols=70):
if len(train[u]) < 1:
continue
if mode == 'valid' and len(valid[u]) < 1:
continue
if mode == 'test' and len(test[u]) < 1:
continue
target_timestamp = None
if args.load_timestamp:
if mode == 'valid':
target_timestamp = valid[u][0][1]
elif mode == 'test':
target_timestamp = test[u][0][1]
seq, interval = build_seq(train, valid if mode is 'test' else None, None, u, target_timestamp, args)
# Put target_item_idx in 0-position
target_item_id = valid[u][0] if mode == 'valid' else test[u][0]
if args.load_timestamp:
target_item_id = target_item_id[0]
us.append(u)
seqs.append(seq)
if args.load_timestamp:
intervals.append(interval)
target_item_ids.append(target_item_id)
valid_user += 1
# predictions = model.predict(sess, us, seqs, list(range(1, item_num + 1)), batch_size=batch_size) # (N, M)
# target_item_scores = predictions[np.arange(int(valid_user)), np.array(target_item_ids, dtype=np.int32) - 1] # (N, )
# ranks = np.sum(np.expand_dims(target_item_scores, -1) < predictions, axis=-1)
# NDCG = np.sum((1 / np.log2(ranks + 2)) * (ranks < 10))
# HT = np.sum(ranks < 10)
# print('NDCG: {} | HT: {}'.format(NDCG, HT))
print('Predicting on evaluate data...')
if eval_item_not_in_history:
if len(evaluate_item) > 0:
top_k_values, top_k_indices = predict_top_k_with_item_set(evaluate_item, model, sess, us, seqs,
batch_size,
user_train=train,
user_valid=valid if mode == 'test' else None,
convert_hist_item_id=True)
else:
top_k_values, top_k_indices = model.predict_top_k(sess, us, seqs, list(range(item_num + 1)), batch_size,
train, valid if mode == 'test' else None,
interval=intervals, top_k=200)
elif len(evaluate_item) > 0:
top_k_values, top_k_indices = predict_top_k_with_item_set(evaluate_item, model, sess, us, seqs, batch_size)
else:
top_k_values, top_k_indices = model.predict_top_k(sess, us, seqs, list(range(1, item_num + 1)),
batch_size=batch_size, interval=intervals)
top_k_indices += 1
top_10_indices = top_k_indices[:, :10]
HT = np.sum(np.expand_dims(target_item_ids, axis=-1) == top_10_indices)
NDCG = np.sum(
(np.expand_dims(target_item_ids, axis=-1) == top_10_indices) *
(1 / np.expand_dims(np.log2(np.arange(0, 10) + 2), axis=0))
)
# print('NDCG: {} | HT: {}'.format(NDCG, HT))
HT /= valid_user
NDCG /= valid_user
top_100_indices = top_k_indices[:, :100]
HT100 = np.sum(np.expand_dims(target_item_ids, axis=-1) == top_100_indices)
NDCG100 = np.sum(
(np.expand_dims(target_item_ids, axis=-1) == top_100_indices) *
(1 / np.expand_dims(np.log2(np.arange(0, 100) + 2), axis=0))
)
HT100 /= valid_user
NDCG100 /= valid_user
top_200_indices = top_k_indices[:, :200]
HT200 = np.sum(np.expand_dims(target_item_ids, axis=-1) == top_200_indices)
NDCG200 = np.sum(
(np.expand_dims(target_item_ids, axis=-1) == top_200_indices) *
(1 / np.expand_dims(np.log2(np.arange(0, 200) + 2), axis=0))
)
HT200 /= valid_user
NDCG200 /= valid_user
return NDCG, HT, NDCG100, HT100, NDCG200, HT200