-
Notifications
You must be signed in to change notification settings - Fork 0
/
mnist.py
218 lines (180 loc) · 8.81 KB
/
mnist.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
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
# Copyright 2016 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""Convolutional Neural Network Estimator for MNIST, built with tf.layers."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import tensorflow as tf
class Mnist:
"""Classifier for MNIST"""
def __init__(self, model_folder):
self._model_folder = model_folder
# Build the CNN (without running it!)
# specifying the operation to use to build the model (when data will be actually passed runtime)
# and the folder where to write the generated models.
self._classifier = tf.estimator.Estimator(
model_fn=self.model_function,
model_dir=self._model_folder)
# Set up logging every 50 steps.
# Log the values in the "softmax_tensor" layer (which is the output layer) with label "probabilities"
self._tensors_to_log = {"probabilities": "softmax_tensor"}
self._logging_hook = tf.train.LoggingTensorHook(
tensors=self._tensors_to_log,
every_n_iter=1000)
def model_function(self, features, labels, mode):
"""Model function for CNN."""
# Input Layer
# The input seems to be one giant "features["x"] array of values - ugh, how unconventional.
# Reshape features["x"] to 4-D tensor: [batch_size, width, height, channels]
# MNIST images are 28x28 pixels, and have one color channel
# Note: the batch_size is left as a variable to compute automatically depending on the features["x"] size.
batch = features["x"] # this is (100, 28, 28, 1)
reshaped_batch = tf.reshape(batch, (-1, 28, 28, 1)) # this is (100, 28, 28, 1)
reshape_layer = tf.placeholder_with_default(reshaped_batch, (None, 28, 28, 1), name="reshape_layer") # this is (?, 28, 28, 1)
# input_layer = tf.reshape(features["x"], [-1, 28, 28, 1], name="input_layer")
# Convolutional Layer #1
# Computes 32 features using a 5x5 filter with ReLU activation.
# the amount of features is the amount of neurons in the layer
# and it corresponds to the "depth" of the layer output.
# Padding is added to preserve width and height.
# Input Tensor Shape: [batch_size, 28, 28, 1]
# Output Tensor Shape: [batch_size, 28, 28, 32]
conv1 = tf.layers.conv2d(
inputs=reshape_layer,
filters=32,
kernel_size=[5, 5],
padding="same",
activation=tf.nn.relu,
name="convolutional_layer_1")
# Pooling Layer #1
# First max pooling layer with a 2x2 filter and stride of 2
# Input Tensor Shape: [batch_size, 28, 28, 32]
# Output Tensor Shape: [batch_size, 14, 14, 32]
pool1 = tf.layers.max_pooling2d(inputs=conv1,
pool_size=[2, 2],
strides=2,
name="pooling_layer_1")
# Convolutional Layer #2
# Computes 64 features using a 5x5 filter.
# Padding is added to preserve width and height.
# Input Tensor Shape: [batch_size, 14, 14, 32]
# Output Tensor Shape: [batch_size, 14, 14, 64]
conv2 = tf.layers.conv2d(
inputs=pool1,
filters=64,
kernel_size=[5, 5],
padding="same",
activation=tf.nn.relu,
name="convolutional_layer_2")
# Pooling Layer #2
# Second max pooling layer with a 2x2 filter and stride of 2
# Input Tensor Shape: [batch_size, 14, 14, 64]
# Output Tensor Shape: [batch_size, 7, 7, 64]
pool2 = tf.layers.max_pooling2d(inputs=conv2,
pool_size=[2, 2],
strides=2,
name="pooling_layer_2")
# Flatten tensor into a batch of vectors
# Input Tensor Shape: [batch_size, 7, 7, 64]
# Output Tensor Shape: [batch_size, 7 * 7 * 64]
pool2_flat = tf.reshape(pool2, [-1, 7 * 7 * 64])
# Dense Layer
# Densely connected layer with 1024 neurons
# Input Tensor Shape: [batch_size, 7 * 7 * 64]
# Output Tensor Shape: [batch_size, 1024]
dense = tf.layers.dense(inputs=pool2_flat,
units=1024,
activation=tf.nn.relu,
name="dense_layer_1")
# Add dropout operation; 0.6 probability that element will be kept
# dropout = tf.layers.dropout(
# inputs=dense, rate=0.4,
# training=mode == tf.estimator.ModeKeys.TRAIN,
# name="dropout_layer")
# Logits layer
# Input Tensor Shape: [batch_size, 1024]
# Output Tensor Shape: [batch_size, 10]
logits = tf.layers.dense(inputs=dense,
units=10,
name="logits_layer")
# tensor with the class chosen from each vector of logits in the batch
# (using one hot encoding)
classes = tf.argmax(input=logits, axis=1)
# tensor containing for each vector of logits in the batch
# the corresponding vector of probabilities computed with a softmax function
probabilities = tf.nn.softmax(logits, name="softmax_tensor")
# Generate predictions in a softmax_tensor layer as a map
# "classes" -> list of classes
# "probabilities" -> list of probability values for those classes
# NOTE: PREDICT and EVAL modes will actually use the predictions directly.
# TRAIN mode will log the predictions every 50 steps
predictions = {"classes": classes, "probabilities": probabilities}
# If we are in PREDICT mode, return the predictions
if mode == tf.estimator.ModeKeys.PREDICT:
return tf.estimator.EstimatorSpec(
mode=tf.estimator.ModeKeys.PREDICT,
predictions=predictions,
export_outputs={'classify': tf.estimator.export.PredictOutput(predictions)})
# Otherwise we are in either TRAIN or EVAL modes, so compute the Loss function for this batch
loss = tf.losses.sparse_softmax_cross_entropy(labels=labels, logits=logits)
# If we are in TRAIN mode, then configure the Training Optimization operation
if mode == tf.estimator.ModeKeys.TRAIN:
optimizer = tf.train.GradientDescentOptimizer(learning_rate=0.001)
train_op = optimizer.minimize(
loss=loss,
global_step=tf.train.get_global_step())
return tf.estimator.EstimatorSpec(mode=mode,
loss=loss,
train_op=train_op)
# Else we are in EVAL mode, so just add evaluation metrics
eval_metric_ops = {
"accuracy": tf.metrics.accuracy(
labels=labels,
predictions=classes)}
return tf.estimator.EstimatorSpec(mode=mode,
loss=loss,
eval_metric_ops=eval_metric_ops)
def train(self, train_data, train_labels, steps=20000):
"""Train the model"""
train_input_fn = tf.estimator.inputs.numpy_input_fn(
x={"x": train_data},
y=train_labels,
batch_size=100,
num_epochs=None,
shuffle=True)
self._classifier.train(
input_fn=train_input_fn,
steps=steps,
hooks=[self._logging_hook])
def evaluate(self, eval_data, eval_labels):
"""Evaluate the model"""
# Evaluate the model
eval_input_fn = tf.estimator.inputs.numpy_input_fn(
x={"x": eval_data},
y=eval_labels,
batch_size=50,
num_epochs=1,
shuffle=False)
eval_results = self._classifier.evaluate(
input_fn=eval_input_fn)
return eval_results
def predict(self, sample):
"""Use the model to predict the class of a sample"""
predict_input_fn = tf.estimator.inputs.numpy_input_fn(
x={"x": sample},
shuffle=False
)
results = self._classifier.predict(
input_fn=predict_input_fn)
return results