20220901 developer call notes

[hammer]

20220901

Pre-notes

• Notes from previous meeting
• R: What’s the plan for the performance analysis - is this something I could help with?

PRs

• sgkit PR#891 (Tim) Pedigree inbreeding
• sgkit PR#892 (Tsz-Fung Chan) Add mean imputation function
• sgkit PR#895 (Tim) Replace additive_relationships with pedigree_relationship
• sgkit PR#899 (Tom) Fix max str len for dask

Issues

• sgkit#893 Optional genomic-scale for popgen stats

Discussions

Other topics

Notes

Attendees

1. Jeff Hammerbacher
2. Tom White
3. Rafal Wojdyla

Discussion

• Performance analysis
  • Get familiar with Cubed and Modal
  • Tom has some work to do getting sgkit using Cubed
    • R:
      • Q1: does it make sense to do it right now?
      • Q2: Nvidia blog?
  • Tom Nicholas proposing generalization of chunking API
  • Tom W can give a presentation on Cubed soon
  • map_overlap from Dask could be hard in Cubed
  • GPU stuff and REGENIE probably won’t work in Cubed now
• sgkit PR#891
  • Good to go, but rebase needed
• sgkit PR#892
  • Jerome’s “not convinced about the general approach of creating new _imputed variables.”
• sgkit PR#895
  • Good to go, but rebase needed
• sgkit PR#899
  • A bit confusing what’s going on
  • Will try to add a test
• sgkit#893
  • Tom will put PR together to move forward

[tomwhite]

Relevant xarray issue for generalizing chunking backend: pydata/xarray#6807

[timothymillar]

Thanks for recording these, it's quite helpful seeing the outline of what's been discussed