SUwonglab/PECA

Request for code clean-up and question about init. of "EM-like" approach

Opened this issue · 2 comments

t-silvers commented

Hi, Thanks for the nice approach and follow ups TimeReg and vPECA.

  1. Gentle request for code clean-up: It would help open source development for some minor clean-up (var names, comments, typo fixes, ...). Then other devs could more easily work on larger refactorings, e.g. the abstraction of PECA/scr/PECA_network_* functions and so on.
  2. Can you describe the initialization strategy for your "EM-like" approach? For instance, I see here you fix eita (eta, $\eta$ ?) to two values:
eita0=-30.4395;
eita1=0.8759;

I'm surprised single values are fixed and fixed to 4 decimal places for this algorithm.

t-silvers commented

Hi @durenzn, bumping this for when you have the chance to take a look.

durenzn commented

Thanks. The eita0 and eita1 parameters were learned from "EM-like" approach PECA model. This is the code for PECA2, which do inference by refining parameters based on the pre-trained PECA model. Those four decimal values are from the training of PECA model for hundreds of cellular context.