DANCo Fast

[njcprice]

Hi,

I noticed in the DANCo.py file that some functions have been commented out which seem to implement a version of FastDANCo. I was wondering if it's safe to uncomment and use these in my code or whether there was a reason as to why they're commented out?

Thanks.

P.S. Great package.

[j-bac]

Hi, thanks for using the package !

That is correct, this is a draft for FastDANCo/DANCoFit, which should work but is pending some validation tests, so I kept it commented out.

It is probably best if you generate your own precomputed interpolation files, this way you can pre-train FastDANCo according to your preferred range and coarse-graining of cardinalities and maximum intrinsic dimension.

I reuploaded for you the file skdim.id.DANCoUtils which contains the code to precompute splines. E.g., to train splines for cardinalities between [100, 500] and intrinsic dimension up to 100:

dHat,mu,tau,DANCo_splines = skdim.id.DANCoUtils.DANCoTrain(k=10, D=100, cardinalities=[100,200,500]), iterations=100, save_output_path='')
dHat,mu,tau,DANCo_splines = skdim.id.DANCoUtils.load_DANCoTrain_results(path='')
skdim.id.DANCoUtils.plot_DANCoTrain([100,200,500],dHat,mu,tau,DANCo_splines)

DANCoTrain is the essential line, it will generate 3 .pkl files which you can put in the skdim.id.DANCoFit/ folder

Then you can uncomment lines 205 - 312 and 361 - 367 in skdim.id.DANCo and you should be able to use FastDANCo with

skdim.id.DANCo(ver='DANCoFit')

[njcprice]

Thanks so much, man. That's really appreciated.