Locally-connected layer for 1D inputs.
LocallyConnected1D layer works similarly to
Conv1D layer, except that weights are unshared,
that is, a different set of filters is applied at each different patch
of the input.
# apply a unshared weight convolution 1d of length 3 to a sequence with # 10 timesteps, with 64 output filters model = Sequential() model.add(LocallyConnected1D(64, 3, input_shape=(10, 32))) # now model.output_shape == (None, 8, 64) # add a new conv1d on top model.add(LocallyConnected1D(32, 3)) # now model.output_shape == (None, 6, 32)
filters: Integer, the dimensionality of the output space (i.e. the number of output filters in the convolution).
kernel_size: An integer or tuple/list of a single integer, specifying the length of the 1D convolution window.
strides: An integer or tuple/list of a single integer, specifying the stride length of the convolution. Specifying any stride value != 1 is incompatible with specifying any
dilation_ratevalue != 1.
padding: Currently only supports
"same"may be supported in the future.
data_format: A string, one of
channels_first. The ordering of the dimensions in the inputs.
channels_lastcorresponds to inputs with shape
(batch, length, channels)while
channels_firstcorresponds to inputs with shape
(batch, channels, length). It defaults to the
image_data_formatvalue found in your Keras config file at
~/.keras/keras.json. If you never set it, then it will be "channels_last".
activation: Activation function to use. If you don't specify anything, no activation is applied (ie. "linear" activation:
a(x) = x).
use_bias: Boolean, whether the layer uses a bias vector.
kernel_initializer: Initializer for the
bias_initializer: Initializer for the bias vector.
kernel_regularizer: Regularizer function applied to the
bias_regularizer: Regularizer function applied to the bias vector.
activity_regularizer: Regularizer function applied to the output of the layer (its "activation")..
kernel_constraint: Constraint function applied to the kernel matrix.
bias_constraint: Constraint function applied to the bias vector.
implementation: implementation mode, either
1loops over input spatial locations to perform the forward pass. It is memory-efficient but performs a lot of (small) ops.
2stores layer weights in a dense but sparsely-populated 2D matrix and implements the forward pass as a single matrix-multiply. It uses a lot of RAM but performs few (large) ops.
3stores layer weights in a sparse tensor and implements the forward pass as a single sparse matrix-multiply.
How to choose:
1: large, dense models,
2: small models,
3: large, sparse models,
where "large" stands for large input/output activations (i.e. many
output_size), and "sparse" stands for few connections between inputs and outputs, i.e. small ratio
filters * input_filters * kernel_size / (input_size * strides), where inputs to and outputs of the layer are assumed to have shapes
It is recommended to benchmark each in the setting of interest to pick the most efficient one (in terms of speed and memory usage). Correct choice of implementation can lead to dramatic speed improvements (e.g. 50X), potentially at the expense of RAM.
padding="valid"is supported by
3D tensor with shape:
(batch_size, steps, input_dim)
3D tensor with shape:
(batch_size, new_steps, filters)
steps value might have changed due to padding or strides.
__init__( filters, kernel_size, strides=1, padding='valid', data_format=None, activation=None, use_bias=True, kernel_initializer='glorot_uniform', bias_initializer='zeros', kernel_regularizer=None, bias_regularizer=None, activity_regularizer=None, kernel_constraint=None, bias_constraint=None, implementation=1, **kwargs )