-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathexp.py
191 lines (173 loc) · 6.53 KB
/
exp.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
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
import os
import cornac
from cornac.data import Reader
from eval_method import QuestERStratifiedSplit
from text_modality import ReviewAndItemQAModality
from quester import QuestER
from cornac.data.text import BaseTokenizer
import numpy as np
import tensorflow as tf
physical_devices = tf.config.list_physical_devices("GPU")
try:
tf.config.experimental.set_memory_growth(physical_devices[0], True)
except:
# Invalid device or cannot modify virtual devices once initialized.
pass
def parse_arguments():
import argparse
parser = argparse.ArgumentParser()
parser.add_argument("-i", "--input", type=str, help="input directory")
parser.add_argument("-ct", "--cluster_threshold", type=float, default=0.8)
parser.add_argument("-mu", "--min_user_freq", type=int, default=5)
parser.add_argument("-mi", "--min_item_freq", type=int, default=5)
parser.add_argument("-na", "--max_num_answer", type=int, default=1)
parser.add_argument("-k", "--n_factors", type=int, default=8)
parser.add_argument("-e", "--epoch", type=int, default=20)
parser.add_argument("-bs", "--batch_size", type=int, default=64)
parser.add_argument(
"-s", "--model_selection", type=str, choices=["best", "last"], default="best"
)
parser.add_argument("-lr", "--learning_rate", type=float, default=0.001)
return parser.parse_args()
args = parse_arguments()
feedback = Reader(min_user_freq=args.min_user_freq, min_item_freq=args.min_item_freq).read(os.path.join(args.input, "rating.txt"), fmt="UIRT", sep="\t")
reviews = Reader().read(
os.path.join(args.input, "review.txt"), fmt="UIReview", sep="\t"
)
data_dir = args.input
MAX_VOCAB = 4000
EMB_SIZE = 100
ID_EMB_SIZE = args.n_factors
N_FACTORS = args.n_factors
ATTENTION_SIZE = 8
BATCH_SIZE = args.batch_size
MAX_NUM_REVIEW = 32
MAX_NUM_QUESTION = 32
MAX_NUM_ANSWER = args.max_num_answer
MAX_TEXT_LENGTH = 128
DROPOUT_RATE = 0.5
TEST_SIZE = 0.1
VAL_SIZE = 0.1
KERNEL_SIZES = [3]
N_FILTERS = 64
CLUSTER_THRESHOLD = 0.8
centroid_questions_file = open(os.path.join(data_dir, "centroid_questions.txt"), "r")
centroid_questions = centroid_questions_file.readlines()
cluster_label_in_order = []
cluster_count = []
with open(os.path.join(data_dir, "cluster.count"), "r") as f:
for line in f:
tokens = line.split(",")
cluster_label_in_order.append(int(tokens[0]))
cluster_count.append(int(tokens[1]))
pct = np.array(cluster_count) / sum(cluster_count)
max_keep_idx = 0
for i in range(len(pct)):
if pct[: i + 1].sum() >= args.cluster_threshold:
max_keep_idx = i + 1
break
print("Max keep idx (coverage:{}): {}".format(args.cluster_threshold, max_keep_idx))
item_question_clusters = {}
with open(os.path.join(data_dir, "item_question_clusters.txt"), "r") as f:
for line in f:
tokens = line.split(",")
item_question_clusters[tokens[0]] = [int(cluster) for cluster in tokens[1:]]
qas = []
with open(os.path.join(data_dir, "qa.txt"), "r") as f:
for line in f:
tokens = line.split("\t\t")
asin = tokens[0]
qas.append(
(
asin,
[
tuple(
[
qtoken
for q_inc, qtoken in enumerate(question.split("\t"))
if q_inc % 2 == 0
]
)
for question, cluster_label in zip(
tokens[1:], item_question_clusters.get(asin, [])
)
if cluster_label in cluster_label_in_order[:max_keep_idx]
],
)
)
mean_question = " ".join(centroid_questions[max_keep_idx:]).replace("\n", " ")
item_with_qas = [x[0] for x in qas]
item_without_qas = list(set([x[1] for x in feedback if x[1] not in item_with_qas]))
[x[1].append((mean_question,)) for x in qas]
qas = qas + [(x, [(mean_question,)]) for x in item_without_qas]
review_and_item_qa_modality = ReviewAndItemQAModality(
data=reviews,
qa_data=qas,
tokenizer=BaseTokenizer(stop_words="english"),
max_vocab=MAX_VOCAB,
)
eval_method = QuestERStratifiedSplit(
data=feedback,
group_by="item",
test_size=TEST_SIZE,
val_size=VAL_SIZE,
exclude_unknowns=True,
review_and_item_qa_text=review_and_item_qa_modality,
verbose=True,
seed=123,
)
pretrained_word_embeddings = {}
with open(f"download/glove/glove.6B.{EMB_SIZE}d.txt", encoding="utf-8") as f:
for line in f:
values = line.split()
word = values[0]
coefs = np.asarray(values[1:], dtype="float32")
pretrained_word_embeddings[word] = coefs
models = [
QuestER(
# name=f"QuestER",
name=f"{os.path.basename(data_dir)}_QuestER_F_{args.n_factors}_A_{ATTENTION_SIZE}_NReview_{MAX_NUM_REVIEW}_NQuestion_{MAX_NUM_QUESTION}_NAnswer_{MAX_NUM_ANSWER}_E_{args.epoch}_BS_{BATCH_SIZE}",
embedding_size=EMB_SIZE,
id_embedding_size=ID_EMB_SIZE,
n_factors=args.n_factors,
attention_size=ATTENTION_SIZE,
kernel_sizes=KERNEL_SIZES,
n_filters=N_FILTERS,
dropout_rate=DROPOUT_RATE,
max_text_length=MAX_TEXT_LENGTH,
max_num_review=MAX_NUM_REVIEW,
max_num_question=MAX_NUM_QUESTION,
max_num_answer=MAX_NUM_ANSWER,
batch_size=BATCH_SIZE,
max_iter=args.epoch,
model_selection=args.model_selection,
optimizer="adam",
learning_rate=args.learning_rate,
init_params={"pretrained_word_embeddings": pretrained_word_embeddings},
verbose=True,
seed=123,
)
]
exp = cornac.Experiment(
eval_method=eval_method,
models=models,
metrics=[
cornac.metrics.MSE(),
],
)
exp.run()
print(data_dir)
selected_model = models[0]
epoch = selected_model.best_epoch if args.model_selection == 'best' else args.epochs.split(',')[0]
model_name = '{}_e_{}'.format(selected_model.name, epoch)
export_dir = os.path.join(args.input, model_name)
os.makedirs(export_dir, exist_ok=True)
import util
from importlib import reload
if args.model_selection == 'best':
util.export_ranked_questions(selected_model, os.path.join(export_dir, 'ranked_questions.txt'))
util.export_useful_review_ranking(selected_model, os.path.join(export_dir, 'useful_review_ranking.txt'))
util.export_most_useful_review(selected_model, os.path.join(export_dir, 'most_useful_review.txt'))
util.export_important_question_ranking(selected_model, os.path.join(export_dir, 'important_question_ranking.txt'))
util.export_quester_explanations(selected_model, export_dir)
# import pdb; pdb.set_trace()