degleris1/cmf.jl

Convolutive Matrix Factorization in Julia

cmf.jl's Issues

Add regularization settings to CMF_results

To be unambiguous, we should save the l1/l2 regularization settings and maybe also the algorithm that was used in CMF_results.

Modify struct `CNMF_results`

Several function take CNMF_results as an argument, including important visualization tools.

However, if the user has a W and H from some other algorithm, they must put in the CNMF_results struct first and set a number of parameters to nothing. I propose we either (a) overload these functions with their corresponding non-struct arguments or (b) change the last fields of the struct (l1_reg, l2_reg, ...) to a single field model_params which contains a dictionary.

plot_Ws broken: eachrow not defined

@ahwillia
I'm having an issue using the sortperm(r::CNMF_results) because eachrow is undefined. My best guess is that you're using Julia 1.1 and that it's a difference between 1.0 and 1.1.

Do we want to change this bit of code to be compatible with 1.0 or do we want to require 1.1 and above?

using CMF broken

after doing a pkg add, using CMF has a load error "ANLS not defined"...

pretty sure this is because include("./anls.jl") is commented out in the CMF module.

Need option to run until convergence

Right now we don't have any convergence checking – we always run until we reach the max time or maximum number of iterations. We should add some simple convergence checks (like just checking when the loss stops changing for example).