def conv2d(h_num_filters,
h_filter_width,
h_stride=1,
h_dilation_rate=1,
h_use_bias=True,
h_padding='SAME'):
def compile_fn(di, dh):
conv_op = tf.layers.Conv2D(dh['num_filters'],
dh['filter_width'],
dh['stride'],
use_bias=dh['use_bias'],
dilation_rate=dh['dilation_rate'],
padding=dh['padding'])
def fn(di):
return {'out': conv_op(di['in'])}
return fn
return siso_tensorflow_module(
'Conv2D', compile_fn, {
'num_filters': h_num_filters,
'filter_width': h_filter_width,
'stride': h_stride,
'use_bias': h_use_bias,
'dilation_rate': h_dilation_rate,
'padding': h_padding
})
def global_pool2d():
def compile_fn(di, dh):
def fn(di):
return {'out': tf.reduce_mean(di['in'], [1, 2])}
return fn
return siso_tensorflow_module('GlobalAveragePool', compile_fn, {})
def fc_layer(h_num_units):
def compile_fn(di, dh):
fc = tf.layers.Dense(dh['num_units'])
def fn(di):
return {'out': fc(di['in'])}
return fn
return siso_tensorflow_module('FCLayer', compile_fn,
{'num_units': h_num_units})
def dropout(h_keep_prob):
def compile_fn(di, dh):
p = tf.placeholder(tf.float32)
def fn(di):
return {'out': tf.nn.dropout(di['in'], p)}
return fn, {p: dh['keep_prob']}, {p: 1.0}
return siso_tensorflow_module('Dropout', compile_fn,
{'keep_prob': h_keep_prob})
def batch_normalization():
def compile_fn(di, dh):
p_var = tf.placeholder(tf.bool)
bn = tf.layers.BatchNormalization()
def fn(di):
return {'out': bn(di['in'], training=p_var)}
return fn, {p_var: 1}, {p_var: 0}
return htf.siso_tensorflow_module('BatchNormalization', compile_fn, {})
def dropout(h_drop_rate):
def compile_fn(di, dh):
p = tf.placeholder(tf.float32)
def fn(di):
return {'out': tf.nn.dropout(di['in'], p)}
return fn, {p: 1.0 - dh['drop_rate']}, {p: 1.0}
return htf.siso_tensorflow_module('Dropout', compile_fn,
{'drop_rate': h_drop_rate})
def avg_pool2d(h_kernel_size, h_stride=1):
def compile_fn(di, dh):
def fn(di):
return {
'out':
tf.nn.avg_pool(di['in'],
[1, dh['kernel_size'], dh['kernel_size'], 1],
[1, dh['stride'], dh['stride'], 1], 'SAME')
}
return fn
return siso_tensorflow_module('MaxPool2D', compile_fn, {
'kernel_size': h_kernel_size,
'stride': h_stride,
})
def nonlinearity(h_nonlin_name):
def compile_fn(di, dh):
def fn(di):
nonlin_name = dh['nonlin_name']
if nonlin_name == 'relu':
Out = tf.nn.relu(di['in'])
elif nonlin_name == 'tanh':
Out = tf.nn.tanh(di['in'])
elif nonlin_name == 'elu':
Out = tf.nn.elu(di['in'])
else:
raise ValueError
return {"out": Out}
return fn
return htf.siso_tensorflow_module('Nonlinearity', compile_fn,
{'nonlin_name': h_nonlin_name})
def relu():
return siso_tensorflow_module(
'ReLU', lambda di, dh: lambda di: {'out': tf.nn.relu(di['in'])}, {})