FluxExtra

Additional layers and functions for the Flux.jl machine learning library.

Layers

Join

Join(dim::Int64)
Join(dim = dim::Int64)

Concatenates a tuple of arrays along a dimension dim. A convenient and type stable way of using x -> cat(x..., dims = dim).

Split

Split(outputs::Int64,dim::Int64)
Split(outputs::Int64, dim = dim::Int64)

Breaks an array into a number of arrays which is equal to output along a dimension dim. dim should we divisible by outputs without a remainder.

Flatten

Flatten()

Flattens an array. A convenient way of using x -> Flux.flatten(x).

Addition

Addition()

A convenient way of using x -> sum(x).

Activation

Activation(f::Function)

A convenient way of using x -> f(x).

Identity

Identity()

Returns its input without changes. Should be used with a Parallel layer if one wants to have a branch that does not change its input.

Normalizations

[0,1]

norm_01!(data::Vector{T}) where {F<:AbstractFloat,N,T<:Array{F,N}}

Rescales each feature (last dimension) to be in the range [0,1]. Returns min and max values for each feature.

norm_01!(data::T,min_vals::T,max_vals::T) where {F<:AbstractFloat,N,T<:Array{F,N}}

Rescales each feature (last dimension) to be in the range [0,1].

[-1,1]

norm_negpos1(data::Vector{T}) where {F<:AbstractFloat,N,T<:Array{F,N}}

Rescales each feature (last dimension) to be in the range [-1,1]. Returns min and max values for each feature.

norm_negpos1(data::T,min_vals::T,max_vals::T) where {F<:AbstractFloat,N,T<:Array{F,N}}

Rescales each feature (last dimension) to be in the range [-1,1].

Zero center

norm_zerocenter!(data::Vector{T}) where {F<:AbstractFloat,N,T<:Array{F,N}}

Subtracts the mean of each feature (last dimension). Returns a mean value for each feature.

norm_zerocenter!(data::T,min_vals::T,max_vals::T) where {F<:AbstractFloat,N,T<:Array{F,N}}

Subtracts the mean of each feature (last dimension).

Z-score

norm_zscore!(data::Vector{T}) where {F<:AbstractFloat,N,T<:Array{F,N}}

Subtracts the mean and divides by the standard deviation of each feature (last dimension). Returns mean and standard deviation values for each feature.

norm_zscore!(data::T,mean_vals::T,std_vals::T) where {F<:AbstractFloat,N,T<:Array{F,N}}

Subtracts the mean and divides by the standard deviation of each feature (last dimension).

Other

Makes Flux.Parallel layer type stable when used with tuples.