-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathgpu_noFeed_v5.pred.py
193 lines (183 loc) · 9.55 KB
/
gpu_noFeed_v5.pred.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
192
193
#!/usr/bin/python
# -*- coding:utf-8 -*-
'''
非feed流下，训练与预测；
多GPU-Test.
加入 learning-decay, tf.train.exponential_decay)
tf.layers.batch_normalization
tf.clip_by_global_norm
'''
import os
import sys
reload(sys)
sys.setdefaultencoding('utf-8')
import time
import numpy as np
import tensorflow as tf
from base import average_gradients
_CSV_COLUMNS = [
'age', 'workclass', 'fnlwgt', 'education', 'education_num',
'marital_status', 'occupation', 'relationship', 'race', 'gender',
'capital_gain', 'capital_loss', 'hours_per_week', 'native_country',
'income_bracket']
_CSV_COLUMN_DEFAULTS = [[0], [''], [0], [''], [0], [''], [''], [''], [''], [''],
[0], [0], [0], [''], ['']]
train_data_file = './data/adult.data'
test_data_file = './data/adult.test'
train_data_file_part = './data/adult.data.head1000'
def get_batch_data(file_name, batch_size=10, buffer_size=100, epoch=10, shuffle=True, drop_last=True, embedding_map=None, reader=None):
columns = build_model_columns(embedding_map, reader)
def decode_line(line):
columns = tf.decode_csv(line, \
record_defaults=_CSV_COLUMN_DEFAULTS)
return dict(zip(_CSV_COLUMNS, columns))
#return columns[:-1], columns[-1:]
## 注意record_defaults会作为默认数据类型去检查field的内容 ##
def tensor_from_input_layer(input, columns):
return tf.feature_column.input_layer( \
features = input, \
feature_columns = columns, \
trainable = True)
## 注意，这个函数在dataset流里处理是快的，不要放到session里面执行，会超级慢 ##
dataset = tf.contrib.data.TextLineDataset(file_name)
dataset = dataset.map(decode_line, num_threads = 5)
if shuffle: dataset = dataset.shuffle(buffer_size = buffer_size)
dataset = dataset.repeat(count = epoch)
if drop_last == True:
dataset = dataset.apply(tf.contrib.data.batch_and_drop_remainder(batch_size))
else:
dataset = dataset.batch(batch_size = batch_size)
dataset = dataset.prefetch(batch_size*2)
dataset = dataset.make_one_shot_iterator()
feature = dataset.get_next()
labels = tf.equal(feature.pop('income_bracket'), '>50K')
labels = tf.reshape(labels, [-1])
transFea= tensor_from_input_layer(feature, columns)
transLab= tf.one_hot(tf.cast(labels, tf.int32), 2)
print 'feature:', feature
print 'transFea:', transFea
print 'labels:', labels
print 'transLab:', transLab
return transFea, transLab
def build_model_columns(embedding_map=None, reader=None):
# Continuous columns
age = tf.feature_column.numeric_column('age')
education_num = tf.feature_column.numeric_column('education_num')
capital_gain = tf.feature_column.numeric_column('capital_gain')
capital_loss = tf.feature_column.numeric_column('capital_loss')
hours_per_week = tf.feature_column.numeric_column('hours_per_week')
education = tf.feature_column.categorical_column_with_vocabulary_list(
'education', [
'Bachelors', 'HS-grad', '11th', 'Masters', '9th', 'Some-college',
'Assoc-acdm', 'Assoc-voc', '7th-8th', 'Doctorate', 'Prof-school',
'5th-6th', '10th', '1st-4th', 'Preschool', '12th'])
marital_status = tf.feature_column.categorical_column_with_vocabulary_list(
'marital_status', [
'Married-civ-spouse', 'Divorced', 'Married-spouse-absent',
'Never-married', 'Separated', 'Married-AF-spouse', 'Widowed'])
relationship = tf.feature_column.categorical_column_with_vocabulary_list(
'relationship', [
'Husband', 'Not-in-family', 'Wife', 'Own-child', 'Unmarried',
'Other-relative'])
workclass = tf.feature_column.categorical_column_with_vocabulary_list(
'workclass', ['?', 'Federal-gov', 'Local-gov', 'Never-worked',
'Private', 'Self-emp-inc', 'Self-emp-not-inc', 'State-gov', 'Without-pay', 'human'])
occupation = tf.feature_column.categorical_column_with_vocabulary_list(
'occupation',['?', 'Adm-clerical', 'Armed-Forces', 'Craft-repair', 'Exec-managerial',
'Farming-fishing', 'Handlers-cleaners', 'Machine-op-inspct', 'Other-service',
'Priv-house-serv', 'Prof-specialty', 'Protective-serv', 'Sales', 'Tech-support', 'Transport-moving'])
age_buckets = tf.feature_column.bucketized_column(
age, boundaries=[18, 25, 30, 35, 40, 45, 50, 55, 60, 65])
em_workclass= tf.feature_column.embedding_column(workclass, dimension=5, combiner='mean', initializer=None)
em_education= tf.feature_column.embedding_column(education, dimension=5, combiner='mean', initializer=None)
em_occupation=tf.feature_column.embedding_column(occupation, dimension=5, combiner='mean', initializer=None)
columns = [
age,
capital_gain,
capital_loss,
hours_per_week,
tf.feature_column.indicator_column(marital_status),
tf.feature_column.indicator_column(relationship),
em_workclass,
em_education,
em_occupation]
return columns
regularizer = tf.contrib.layers.l2_regularizer(0.01)
model_dir = './model_v5'
#in_x, in_y = get_batch_data(train_data_file_part, shuffle=False, batch_size=1000, epoch=1, drop_last=False)
in_x, in_y = get_batch_data(test_data_file, shuffle=False, batch_size=1000, epoch=1, drop_last=False)
training = False
with tf.variable_scope('layer-1', reuse = tf.AUTO_REUSE):
y1 = tf.layers.dense(inputs = in_x, units = 128, use_bias=False, \
activation = tf.nn.relu, \
kernel_regularizer = regularizer, \
bias_regularizer = None)
y1 = tf.layers.batch_normalization(y1, training= training)
with tf.variable_scope('layer-2', reuse = tf.AUTO_REUSE):
y2 = tf.layers.dense(inputs = y1, units = 64, use_bias=False, \
activation = tf.nn.relu, \
kernel_regularizer = regularizer, \
bias_regularizer = None)
y2 = tf.layers.batch_normalization(y2, training= training)
with tf.variable_scope('layer-3', reuse = tf.AUTO_REUSE):
y3 = tf.layers.dense(inputs = y2, units = 2, use_bias=True, \
activation = None, \
kernel_regularizer = regularizer, \
bias_regularizer = regularizer)
logits = y3
prob_all = tf.nn.softmax(logits, 1)
pred_class = tf.argmax(prob_all, 1)
prob_class = tf.reduce_max(prob_all, 1)
prob_class = tf.cast(tf.expand_dims(prob_class, 1), dtype=tf.float32)
pred_class = tf.cast(tf.expand_dims(pred_class, 1), dtype=tf.float32)
real_label = tf.cast(tf.expand_dims(tf.argmax(in_y, 1), 1), dtype=tf.float32)
pred_res = tf.concat([prob_class, pred_class, real_label], 1)
saver = tf.train.Saver()
ckpt_file = tf.train.latest_checkpoint(model_dir)
print 'here ckpt_file:', ckpt_file
if ckpt_file:
config = tf.ConfigProto(gpu_options = tf.GPUOptions(allow_growth=True), device_count = {'GPU':0}, allow_soft_placement = True)
with tf.Session(config=config) as sess:
saver.restore(sess, ckpt_file)
sess.run(tf.tables_initializer())
global_variables = tf.global_variables()
print 'tf.global_variables', global_variables
for i in global_variables: print i.name
from tensorflow.python import pywrap_tensorflow
reader = pywrap_tensorflow.NewCheckpointReader(ckpt_file)
# ## 这里验证了模型里的参数被加载正确 ##
print 'in model_save, layer-1/bn/moving_mean', reader.get_tensor('layer-1/batch_normalization/moving_mean')
print 'in Graph , layer-1/bn/moving_mean', sess.run('layer-1/batch_normalization/moving_mean:0')
print 'in model_save, layer-1/bn/moving_variance', reader.get_tensor('layer-1/batch_normalization/moving_variance')
print 'in Graph , layer-1/bn/moving_variance', sess.run('layer-1/batch_normalization/moving_variance:0')
print 'in model_save, layer-1/bn/beta', reader.get_tensor('layer-1/batch_normalization/beta')
print 'in Graph , layer-1/bn/beta', sess.run('layer-1/batch_normalization/beta:0')
print 'in model_save, layer-1/bn/gamma', reader.get_tensor('layer-1/batch_normalization/gamma')
print 'in Graph , layer-1/bn/gamma', sess.run('layer-1/batch_normalization/gamma:0')
print 'in model_save, layer-2/bn/moving_mean', reader.get_tensor('layer-2/batch_normalization/moving_mean')
print 'in Graph , layer-2/bn/moving_mean', sess.run('layer-2/batch_normalization/moving_mean:0')
print 'in model_save, layer-2/bn/moving_variance', reader.get_tensor('layer-2/batch_normalization/moving_variance')
print 'in Graph layer-2/bn/moving_variance', sess.run('layer-2/batch_normalization/moving_variance:0')
right = 0
all_num = 0
iter_num= 0
while True:
try:
res = sess.run(pred_res)
right += sess.run(tf.reduce_sum(tf.cast(tf.equal(res[:,1], res[:,2]), tf.int32)))
all_num+= res.shape[0]
#print res
except tf.errors.OutOfRangeError:
print 'pred end'
break
if iter_num % 10 == 0: ## 这里查看图中的变量，是一致未有变化的 ##验证 training=False下，BN的学习到的系数都是不动的 ##
# print 'iter:', iter_num, 'in Graph , layer-1/bn/moving_mean', sess.run('layer-1/batch_normalization/moving_mean:0')
# print 'iter:', iter_num, 'in Graph , layer-1/bn/moving_variance', sess.run('layer-1/batch_normalization/moving_variance:0')
# print 'iter:', iter_num, 'in Graph , layer-2/bn/moving_mean', sess.run('layer-2/batch_normalization/moving_mean:0')
# print 'iter:', iter_num, 'in Graph layer-2/bn/moving_variance', sess.run('layer-2/batch_normalization/moving_variance:0')
print 'iter:', iter_num, 'in Graph , layer-1/bn/beta', sess.run('layer-1/batch_normalization/beta:0')
print 'iter:', iter_num, 'in Graph , layer-1/bn/gamma', sess.run('layer-1/batch_normalization/gamma:0')
iter_num += 1
print 'right-num:', right, 'all_num:', all_num, 'accuracy:', float(right)/float(all_num)
else:
print 'no check point in path:', model_dir