kr-colab / dispersenn2 Goto Github PK

Currently the docs (and paper) are a bit sparse about how the empirical locations are treated.

Missing SNPs (not imputed) in the empirical VCF will cause an error

@stsmall had this idea

I've been trying out this method for a few weeks now and I'm curious to see it develop further.

One slight issue I ran into was the input format for the geographic coordinates. By default, disperseNN2 only accepts coordinates expressed in decimal degrees (latitude and longitude), but some of my datasets have UTM coordinates.

In the code, disperseNN2 converts the decimal degrees into UTM anyways (with utm.from_latlong), so it would be nice to have it accept UTM coordinates as input also. It is not a major issue but would spare the user from having to do the conversion to decimal degrees every time.

Keep up the good work! 😄

Right now I parallelize this step by different random number seeds.

Would be nicer to run just cone command with --threads

Currently SNPs are ordered by genomic position to take advantage of LD information. But for reference-free SNPs, or sparse SNPs, users might want to train with shuffled SNPs. This would be an easy feature to add.

Running the vignette is giving different output than that in the docs. Most likely this is due to recent commits changing the code, and we just need to update the docs.