OMLT Layers

Neural network layer classes.

class omlt.neuralnet.layer.ConvLayer2D(input_size, output_size, strides, kernel, *, activation=None, input_index_mapper=None)[source]

Bases: Layer2D

Two-dimensional convolutional layer.

Parameters

input_size (tuple) – the size of the input.
output_size (tuple) – the size of the output..
strides (matrix-like) – stride of the cross-correlation kernel.
kernel (matrix-like) – the cross-correlation kernel.
activation (str or None) – activation function name
input_index_mapper (IndexMapper or None) – map indexes from this layer index to the input layer index size

property kernel: Return the cross-correlation kernel

property kernel_depth: Return the depth of the cross-correlation kernel

property kernel_shape: Return the shape of the cross-correlation kernel

kernel_with_input_indexes(out_d, out_r, out_c)[source]

Returns an iterator over the kernel value and input index for the output at index (out_d, out_r, out_c).

Parameters

out_d (int) – the output depth.
out_r (int) – the output row.
out_c (int) – the output column.

class omlt.neuralnet.layer.DenseLayer(input_size, output_size, weights, biases, *, activation=None, input_index_mapper=None)[source]

Bases: Layer

Dense layer implementing output = activation(dot(input, weights) + biases).

Parameters

input_size (tuple) – the size of the input.
output_size (tuple) – the size of the output.
weight (matrix-like) – the weight matrix.
biases (array-like) – the biases.
activation (str or None) – activation function name
input_index_mapper (IndexMapper or None) – map indexes from this layer index to the input layer index size

property biases: Return the vector of node biases

property weights: Return the matrix of node weights

class omlt.neuralnet.layer.IndexMapper(input_size, output_size)[source]

Bases: object

Map indexes from one layer to the other.

Parameters

input_size (tuple) – the input size
output_size (tuple) – the mapped input layer’s output size

property input_size: Return the size of the input tensor

property output_size: Return the size of the output tensor

class omlt.neuralnet.layer.InputLayer(size)[source]

Bases: Layer

The first layer in any network.

Parameters: size (tuple) – the size of the input.

class omlt.neuralnet.layer.Layer(input_size, output_size, *, activation=None, input_index_mapper=None)[source]

Bases: object

Base layer class.

Parameters

input_size (tuple) – size of the layer input
output_size (tuple) – size of the layer output
activation (str or None) – activation function name
input_index_mapper (IndexMapper or None) – map indexes from this layer index to the input layer index size

property activation: Return the activation function

eval_single_layer(x)[source]

Evaluate the layer at x.

Parameters: x (array-like) – the input tensor. Must have size self.input_size.

property input_index_mapper: Return the index mapper

property input_indexes: Return a list of the input indexes

property input_indexes_with_input_layer_indexes

Return an iterator generating a tuple of local and input indexes.

Local indexes are indexes over the elements of the current layer. Input indexes are indexes over the elements of the previous layer.

property input_size: Return the size of the input tensor

property output_indexes: Return a list of the output indexes

property output_size: Return the size of the output tensor

class omlt.neuralnet.layer.Layer2D(input_size, output_size, strides, *, activation=None, input_index_mapper=None)[source]

Bases: Layer

Abstract two-dimensional layer that downsamples values in a kernel to a single value.

Parameters

input_size (tuple) – the size of the input.
output_size (tuple) – the size of the output.
strides (matrix-like) – stride of the kernel.
activation (str or None) – activation function name
input_index_mapper (IndexMapper or None) – map indexes from this layer index to the input layer index size

get_input_index(out_index, kernel_index)[source]: Returns the input index corresponding to the output at out_index and the kernel index kernel_index.

property kernel_depth: Return the depth of the kernel

kernel_index_with_input_indexes(out_d, out_r, out_c)[source]

Returns an iterator over the index within the kernel and input index for the output at index (out_d, out_r, out_c).

Parameters

out_d (int) – the output depth.
out_r (int) – the output row.
out_c (int) – the output column.

property kernel_shape: Return the shape of the kernel

property strides: Return the stride of the layer

class omlt.neuralnet.layer.PoolingLayer2D(input_size, output_size, strides, pool_func_name, kernel_shape, kernel_depth, *, activation=None, input_index_mapper=None)[source]

Bases: Layer2D

Two-dimensional pooling layer.

Parameters

input_size (tuple) – the size of the input.
output_size (tuple) – the size of the output.
strides (matrix-like) – stride of the kernel.
pool_func (str) – name of function used to pool values in a kernel to a single value.
transpose (bool) – True iff input matrix is accepted in transposed (i.e. column-major) form.
activation (str or None) – activation function name
input_index_mapper (IndexMapper or None) – map indexes from this layer index to the input layer index size

property kernel_depth: Return the depth of the kernel

property kernel_shape: Return the shape of the kernel