Reducing the data size: the Reductors module

class nnfbp.Reductors.IdentityReductor(size)[source]

Bases: nnfbp.Reductors.Reductor

An implementation of a Reductor that performs no reduction at all.

class nnfbp.Reductors.LogSymReductor(size, nLinear=2)[source]

Bases: nnfbp.Reductors.Reductor

An implementation of a Reductor with exponentially growing bin widths, and symmetric bins.

Parameters:nLinear (int) – Number of bins of width 1 before starting exponential growth.’
class nnfbp.Reductors.Reductor(inSize)[source]

Bases: object

Base object of a Reductor, that takes input data and reduces it.

Implementing objects should define outSize, the number of elements after reduction, and a filters numpy.ndarray of size (inSize,outSize), where each row is a basis vector.

Parameters:inSize (int) – Input size of vectors.

Returns actual FBP filters, given the resulting weights of a trained neural network.

Previous topic

A simple projector using PyCUDA: the SimplePyCUDAProjector module

Next topic

Datasets: the DataSet module

This Page