def build_network(self):
"""
Building network for mnist
"""
with tf.name_scope('reshape'):
try:
input_dim = int(math.sqrt(self.x_dim))
except:
print('input dim cannot be sqrt and reshape. input dim: ' +
str(self.x_dim))
logger.debug(
'input dim cannot be sqrt and reshape. input dim: %s',
str(self.x_dim))
raise
x_image = tf.reshape(self.images, [-1, input_dim, input_dim, 1])
with tf.name_scope('conv1'):
w_conv1 = weight_variable([self.conv_size, self.conv_size, 1,
self.channel_1_num])
b_conv1 = bias_variable([self.channel_1_num])
h_conv1 = nni.function_choice(lambda : tf.nn.relu(conv2d(
x_image, w_conv1) + b_conv1), lambda : tf.nn.sigmoid(conv2d
(x_image, w_conv1) + b_conv1), lambda : tf.nn.tanh(conv2d(
x_image, w_conv1) + b_conv1), name='tf.nn.relu')
with tf.name_scope('pool1'):
h_pool1 = nni.function_choice(lambda : max_pool(h_conv1, self.
pool_size), lambda : avg_pool(h_conv1, self.pool_size),
name='max_pool')
with tf.name_scope('conv2'):
w_conv2 = weight_variable([self.conv_size, self.conv_size, self
.channel_1_num, self.channel_2_num])
b_conv2 = bias_variable([self.channel_2_num])
h_conv2 = tf.nn.relu(conv2d(h_pool1, w_conv2) + b_conv2)
with tf.name_scope('pool2'):
h_pool2 = max_pool(h_conv2, self.pool_size)
last_dim = int(input_dim / (self.pool_size * self.pool_size))
with tf.name_scope('fc1'):
w_fc1 = weight_variable([last_dim * last_dim * self.
channel_2_num, self.hidden_size])
b_fc1 = bias_variable([self.hidden_size])
h_pool2_flat = tf.reshape(h_pool2, [-1, last_dim * last_dim * self.
channel_2_num])
h_fc1 = tf.nn.relu(tf.matmul(h_pool2_flat, w_fc1) + b_fc1)
with tf.name_scope('dropout'):
h_fc1_drop = tf.nn.dropout(h_fc1, self.keep_prob)
with tf.name_scope('fc2'):
w_fc2 = weight_variable([self.hidden_size, self.y_dim])
b_fc2 = bias_variable([self.y_dim])
y_conv = tf.matmul(h_fc1_drop, w_fc2) + b_fc2
with tf.name_scope('loss'):
cross_entropy = tf.reduce_mean(tf.nn.
softmax_cross_entropy_with_logits(labels=self.labels,
logits=y_conv))
with tf.name_scope('adam_optimizer'):
self.train_step = tf.train.AdamOptimizer(self.learning_rate
).minimize(cross_entropy)
with tf.name_scope('accuracy'):
correct_prediction = tf.equal(tf.argmax(y_conv, 1), tf.argmax(
self.labels, 1))
self.accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.
float32))
def test_default_name_func(self):
val = nni.function_choice({
'max(1, 2, 3)': lambda: max(1, 2, 3),
'min(1, 2)':
lambda: min(1, 2) # NOTE: assign this line number to lineno2
})
self.assertEqual(val, 3)
def test_lambda_func(self):
val = nni.function_choice(
{
"lambda: 2*3": lambda: 2 * 3,
"lambda: 3*4": lambda: 3 * 4
},
name='lambda_func')
self.assertEqual(val, 6)
def test_func(self):
val = nni.function_choice({
'foo': foo,
'bar': bar
},
name='func',
key='test_smartparam/func/function_choice')
self.assertEqual(val, 'bar')
def test_lambda_func(self):
val = nni.function_choice(
{
"lambda: 2*3": lambda: 2 * 3,
"lambda: 3*4": lambda: 3 * 4
},
name='lambda_func',
key='test_smartparam/lambda_func/function_choice')
self.assertEqual(val, 6)
def test_default_name_func(self):
val = nni.function_choice(
lambda: max(1, 2, 3),
lambda: 2 * 2 # NOTE: assign this line number to lineno2
)
self.assertEqual(val, 3)
def test_specified_name_func(self):
val = nni.function_choice(foo, bar, name='func')
self.assertEqual(val, 'bar')
import nni
def max_pool(k):
pass
h_conv1 = 1
conv_size = nni.choice(2, 3, 5, 7, name='conv_size')
h_pool1 = nni.function_choice(lambda: max_pool(h_conv1),
lambda: avg_pool(h_conv2, h_conv3),
name='max_pool')
test_acc = 1
nni.report_intermediate_result(test_acc)
test_acc = 2
nni.report_final_result(test_acc)
import nni
def max_pool(k):
pass
h_conv1 = 1
nni.choice({'foo': foo, 'bar': bar})(1)
conv_size = nni.choice({2: 2, 3: 3, 5: 5, 7: 7}, name='conv_size')
abc = nni.choice({'2': '2', 3: 3, '(5 * 6)': 5 * 6, 7: 7}, name='abc')
h_pool1 = nni.function_choice({
'max_pool': lambda: max_pool(h_conv1),
'h_conv1': lambda: h_conv1,
'avg_pool': lambda: avg_pool(h_conv2, h_conv3)
})
h_pool1 = nni.function_choice(
{
'max_pool(h_conv1)': lambda: max_pool(h_conv1),
'avg_pool(h_conv2, h_conv3)': lambda: avg_pool(h_conv2, h_conv3)
},
name='max_pool')
h_pool2 = nni.function_choice(
{
'max_poo(h_conv1)': lambda: max_poo(h_conv1),
'(2 * 3 + 4)': lambda: 2 * 3 + 4,
'(lambda x: 1 + x)': lambda: lambda x: 1 + x
},
name='max_poo')
tmp = nni.qlognormal(1.2, 3, 4.5)
test_acc = 1
def build_network(self):
'''
Building network for mnist
'''
# Reshape to use within a convolutional neural net.
# Last dimension is for "features" - there is only one here, since images are
# grayscale -- it would be 3 for an RGB image, 4 for RGBA, etc.
with tf.name_scope('reshape'):
try:
input_dim = int(math.sqrt(self.x_dim))
except:
print('input dim cannot be sqrt and reshape. input dim: ' +
str(self.x_dim))
logger.debug(
'input dim cannot be sqrt and reshape. input dim: %s',
str(self.x_dim))
raise
x_image = tf.reshape(self.images, [-1, input_dim, input_dim, 1])
# First convolutional layer - maps one grayscale image to 32 feature maps.
with tf.name_scope('conv1'):
w_conv1 = weight_variable(
[self.conv_size, self.conv_size, 1, self.channel_1_num])
b_conv1 = bias_variable([self.channel_1_num])
h_conv1 = nni.function_choice(
lambda: tf.nn.relu(conv2d(x_image, w_conv1) + b_conv1),
lambda: tf.nn.sigmoid(conv2d(x_image, w_conv1) + b_conv1),
lambda: tf.nn.tanh(conv2d(x_image, w_conv1) + b_conv1
)) # example: without name
# Pooling layer - downsamples by 2X.
with tf.name_scope('pool1'):
h_pool1 = max_pool(h_conv1, self.pool_size)
h_pool1 = nni.function_choice(
lambda: max_pool(h_conv1, self.pool_size),
lambda: avg_pool(h_conv1, self.pool_size),
name='h_pool1')
# Second convolutional layer -- maps 32 feature maps to 64.
with tf.name_scope('conv2'):
w_conv2 = weight_variable([
self.conv_size, self.conv_size, self.channel_1_num,
self.channel_2_num
])
b_conv2 = bias_variable([self.channel_2_num])
h_conv2 = tf.nn.relu(conv2d(h_pool1, w_conv2) + b_conv2)
# Second pooling layer.
with tf.name_scope('pool2'): # example: another style
h_pool2 = max_pool(h_conv2, self.pool_size)
# Fully connected layer 1 -- after 2 round of downsampling, our 28x28 image
# is down to 7x7x64 feature maps -- maps this to 1024 features.
last_dim = int(input_dim / (self.pool_size * self.pool_size))
with tf.name_scope('fc1'):
w_fc1 = weight_variable(
[last_dim * last_dim * self.channel_2_num, self.hidden_size])
b_fc1 = bias_variable([self.hidden_size])
h_pool2_flat = tf.reshape(
h_pool2, [-1, last_dim * last_dim * self.channel_2_num])
h_fc1 = tf.nn.relu(tf.matmul(h_pool2_flat, w_fc1) + b_fc1)
# Dropout - controls the complexity of the model, prevents co-adaptation of features.
with tf.name_scope('dropout'):
h_fc1_drop = tf.nn.dropout(h_fc1, self.keep_prob)
# Map the 1024 features to 10 classes, one for each digit
with tf.name_scope('fc2'):
w_fc2 = weight_variable([self.hidden_size, self.y_dim])
b_fc2 = bias_variable([self.y_dim])
y_conv = tf.matmul(h_fc1_drop, w_fc2) + b_fc2
with tf.name_scope('loss'):
cross_entropy = tf.reduce_mean(
tf.nn.softmax_cross_entropy_with_logits(labels=self.labels,
logits=y_conv))
with tf.name_scope('adam_optimizer'):
self.train_step = tf.train.AdamOptimizer(
self.learning_rate).minimize(cross_entropy)
with tf.name_scope('accuracy'):
correct_prediction = tf.equal(tf.argmax(y_conv, 1),
tf.argmax(self.labels, 1))
self.accuracy = tf.reduce_mean(
tf.cast(correct_prediction, tf.float32))
