To install this package and add it to your primary Julia development environment, run:

$ git clone https://github.com/degleris1/ConvolutionalFactorizations.jl.git
$ cd ConvolutionalFactorizations.jl
$ julia

From the Julia REPL, press ] to enter package mode, and run:

pkg> dev .

Then press [delete] to exit package mode. Alternatively, if you're not a developer, all you need to do is open up the Julia REPL, press ] to enter package mode, and run:

pkg> add https://github.com/degleris1/ConvolutionalFactorizations.jl.git

• One line between functions
• Use the following to delimit different sections of a file:

###
# SECTION NAME
###

• 4 lines to separate a section

Minor

• Performance optimize MaskedLoss.
• Performance optimize penalty functions and constraints.
• Implement transform and update
• Check out jax style optimizers

struct StochasticGradientDescent <: AbstractOptimizer
    learning_rate
end

function initialize(opt::StochasticGradientDescent, params)
    W, H = params

    return (W=W, H=H, )
end


params = (W0, H0)

opt = StochasticGradientDescent(learning_rate)
opt_state = initialize(opt, params)

while !converged(opt, opt_state)
    opt_state = update(opt, opt_state)
end

• In the long term, it would be nice to have an AbstractTensor class with methods like get_slice(W::AbstractTensor, lag::Int) and get_factor(W::AbstractTensor, k::Int). The reason this would be helpful is that different algorithms perform better with different memory layouts, so dispatching on the memory layout could greatly improve performance.