keraflow.layers.convolution.Convolution1D Class Reference

Convolution layer for convolving (sequence_length, input_dim) inputs. More...

Inheritance diagram for keraflow.layers.convolution.Convolution1D:

Public Member Functions
def	__init__
Public Member Functions inherited from keraflow.layers.convolution.ConvolutionBase
def	__init__

Detailed Description

Convolution layer for convolving (sequence_length, input_dim) inputs.

• input_shape: 3D, (nb_samples, sequence_length, input_dim)
• output_shape: 3D, (nb_samples, output_sequence_length, nb_kernel)
• parameters:
  • W: (1, 1, kernel_row, input_col)
  • b: (nb_kernel,)

Note

1. output_sequence_length is determined by pad and stride. For details, please see ConvolutionBase.
2. The shape of W has two additional dimensions (the 1s) due to implementation issue. Be aware of this when initializing the layer with initial_weights argument.

Constructor & Destructor Documentation

def keraflow.layers.convolution.Convolution1D.__init__	(		self,
			nb_kernel,
			kernel_row,
			stride = 1,
			pad = 'valid',
			bias = True,
			init = 'uniform',
			activation = 'linear',
			kwargs
	)

Parameters

nb_kernel	int. Number of convolution kernels to use.
kernel_row	int. The height of the each kernel. The width of kernel will always be the input's width.
stride	int. Step for vertically sliding each kernel for convolution.
pad	str, 'valid' of 'same'. See ConvolutionBase
bias	boolean. Whether to include a bias (i.e. make the layer affine rather than linear).
init	str/function. Function to initialize trainable parameters. See Initializations.
activation	str/function. Activation function applied on the output. See Activations.
kwargs	see Layer.__init__.