keraflow.layers.convolution.PoolingBase Class Reference

Base layer for pooling layers. More...

Inheritance diagram for keraflow.layers.convolution.PoolingBase:

Public Member Functions
def	__init__

Detailed Description

Base layer for pooling layers.

Do not use this layer in your code.

Constructor & Destructor Documentation

def keraflow.layers.convolution.PoolingBase.__init__	(		self,
			mode,
			pool_size,
			strides = None,
			pad = 'valid',
			kwargs
	)

Parameters

mode	str. 'max' or 'avg' for max and average pooling respectively.
pool_size	tuple of int. Shape of the pooling window.
strides	tuple of int. Steps for sliding the pooling window.
pad	str, 'valid' of 'same'. See descriptions below.
kwargs	see Layer.__init__.

Note

The output shape and paddings of pooling ayers is calculated as described in ConvolutionBase.__init__. However, when theano is used as backend, we are unable to conduct tensorflow's asymmetric padding. Current solution is to perform symmetric padding with larger (out of front pad and back pad) padding and truncate the output to match the output shape (determined by the formula).