-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathchar_input_fn.py
174 lines (144 loc) · 7.53 KB
/
char_input_fn.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
#!/usr/bin/env python
# coding=utf-8
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import attrdict
import copy
import datetime
import numpy as np
import os
import re
import tensorflow as tf
from tensorflow.estimator import ModeKeys
import random
import json
import glob
import sys
# distributed
#sys.path.insert(0,'/export/App/training_platform/PinoModel/external/tensorflow/tensorflow/core/custom_ops/python/')
sys.path.insert(0,'/export/sdb/liuziyang7/char_match_gnn/env/tensorflow/tensorflow/core/custom_ops/python/')
#sys.path.insert(0,'/home/admin/chengzhaomeng/tool/stone_Python_3_6_5/bin/python3.6')
import tokenize_fn
class DataConfig(object):
def __init__(self, config=None):
#self.train_dataset_files = "/export/sdb/liuziyang7/char_match_gnn/data/char-gnn-train.tsv"
#self.eval_dataset_files = "/export/sdb/liuziyang7/char_match_gnn/data/char-gnn-eval.tsv"
#self.test_dataset_files = "/export/sdb/liuziyang7/char_match_gnn/data/char-gnn-eval.tsv"
self.train_dataset_files = "/export/sdb/liuziyang7/char_match_gnn/data_170m/train_nei.tsv"
self.eval_dataset_files = "/export/sdb/liuziyang7/char_match_gnn/data_170m/eval_nei.tsv"
self.test_dataset_files = "/export/sdb/liuziyang7/char_match_gnn/data_170m/eval_nei.tsv"
self.vocab_word = "vocab.word"
self.filter_word = "filter.word"
self.filter_list = [ l.strip() for l in open(self.filter_word, 'r').readlines()]
self.filter_dict = dict(zip(self.filter_list, [1 for i in range(len(self.filter_list))]))
self.filter_colsize = 11
self.batch_size = 1024
self.num_epochs = 20
self.perform_shuffle = False
self.num_word_ids = 40000
self.query_size = 10
self.title_size = 65
self.padding_string = "0_0"
data_config = DataConfig()
feature_names = ["q", "sna", "qiq1_t", "qiq1_q", "qiq2_t", "qiq2_q", "iqi1_q", "iqi1_t", "iqi2_q", "iqi2_t"]
def unigram_and_padding(string_tensor, width, padding_value):
sparse_tensor = tokenize_fn.unigrams_alphanum_lower_parser(string_tensor)
return sparse_tensor
def is_in_filter(x):
if x.decode('utf8') in data_config.filter_dict:
#print("1:", x.decode('utf8'))
return np.int32(1)
else:
#print("0:", x.decode('utf8'))
return np.int32(0)
def filter_line(line):
columns = tf.string_split([line], '\t')
column_size = tf.size(columns)
# 1: exist else 0
query_exists = tf.py_func(is_in_filter, [columns.values[0]], tf.int32)
#return tf.equal(query_exists, 0)
return tf.logical_and(tf.equal(query_exists, 0), tf.equal(column_size, data_config.filter_colsize))
def input_fn(filenames, batch_size, num_epochs, perform_shuffle, is_training=False):
def decode_line(line):
columns = tf.string_split([line], '\t')
labels = tf.string_to_number(columns.values[2], out_type=tf.float32)
query_ids = unigram_and_padding(columns.values[0], data_config.query_size, data_config.padding_string)
title_ids = unigram_and_padding(columns.values[1], data_config.title_size, data_config.padding_string)
q_i_q1_title_ids = unigram_and_padding(columns.values[3], data_config.title_size, data_config.padding_string)
q_i_q1_query_ids = unigram_and_padding(columns.values[4], data_config.query_size, data_config.padding_string)
q_i_q2_title_ids = unigram_and_padding(columns.values[5], data_config.title_size, data_config.padding_string)
q_i_q2_query_ids = unigram_and_padding(columns.values[6], data_config.query_size, data_config.padding_string)
i_q_i1_title_ids = unigram_and_padding(columns.values[7], data_config.title_size, data_config.padding_string)
i_q_i1_query_ids = unigram_and_padding(columns.values[8], data_config.query_size, data_config.padding_string)
i_q_i2_title_ids = unigram_and_padding(columns.values[9], data_config.title_size, data_config.padding_string)
i_q_i2_query_ids = unigram_and_padding(columns.values[10], data_config.query_size, data_config.padding_string)
return dict(zip(feature_names, [query_ids, title_ids, q_i_q1_title_ids, q_i_q1_query_ids, q_i_q2_title_ids, q_i_q2_query_ids, i_q_i1_title_ids, i_q_i1_query_ids, i_q_i2_title_ids, i_q_i2_query_ids])), labels
# Extract lines from input files using the Dataset API, can pass one filename or filename list
dataset = tf.data.TextLineDataset(filenames)
if is_training:
dataset = dataset.filter(filter_line)
dataset = dataset.map(decode_line, num_parallel_calls=20).prefetch(50000)
# Randomizes input using a window of 256 elements (read into memory)
if perform_shuffle:
dataset = dataset.shuffle(buffer_size=batch_size*20)
# epochs from blending together.
dataset = dataset.repeat(num_epochs)
dataset = dataset.batch(batch_size) # Batch size to use
return dataset
def get_all_files(filedir):
if filedir.startswith("hdfs"):
if tf.gfile.IsDirectory(filedir):
return [filedir+'/'+ele for ele in tf.gfile.ListDirectory(filedir)]
else:
return [filedir]
else:
return glob.glob("%s" % filedir)
def train_input_fn():
filenames = get_all_files(data_config.train_dataset_files)
return input_fn(filenames, data_config.batch_size, data_config.num_epochs, data_config.perform_shuffle, True)
def eval_input_fn():
filenames = get_all_files(data_config.eval_dataset_files)
return input_fn(filenames, data_config.batch_size, 1, False)
def predict_input_fn():
filenames = get_all_files(data_config.test_dataset_files)
return input_fn(filenames, 1, 1, False)
def export_input_fn():
export_columns = [tf.VarLenFeature(tf.string), tf.VarLenFeature(tf.string)]
result = dict(zip(feature_names, export_columns))
return result
def batch_process_mapper(features, config=None):
for fkey in feature_names:
untokenizer_tensor = features[fkey]
if isinstance(untokenizer_tensor, tf.SparseTensor):
untokenizer_tensor = untokenizer_tensor.values
if fkey == "q":
features[fkey] = unigram_and_padding(untokenizer_tensor, data_config.query_size, data_config.padding_string)
elif fkey == "sna":
features[fkey] = unigram_and_padding(untokenizer_tensor, data_config.title_size, data_config.padding_string)
return features
def word2ids(text):
## define vocabulary
vocabulary_feature_column =tf.feature_column.categorical_column_with_vocabulary_file(key="wordstring",
vocabulary_file=data_config.vocab_word,
vocabulary_size=None)
vocab_len = len(open(data_config.vocab_word, 'r').readlines())
column = tf.feature_column.embedding_column(vocabulary_feature_column, 1, initializer=tf.constant_initializer(np.array([[i] for i in range(vocab_len)])), trainable=False)
## map text into ids
text_str = {"wordstring": tf.reshape(text, [-1])}
text_ids = tf.cast(tf.feature_column.input_layer(text_str, column), dtype=tf.int32)
return text_ids
if __name__ == '__main__':
os.environ['CUDA_VISIBLE_DEVICES'] = '3'
x, y = predict_input_fn().make_one_shot_iterator().get_next()
print("x", x)
print("y", y)
z_1, z_2 = word2ids(x["q"].values, x["sna"].values)
idx = tf.SparseTensor(indices=x["q"].indices,
values=tf.reshape(z_1, [-1]), dense_shape=x["q"].dense_shape)
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
sess.run(tf.tables_initializer())
print(sess.run([x, y]))
print(sess.run([z_1, z_2]))
print(sess.run(idx))