Pythonic API

[flying-sheep]

This is a straight translation from R. I’d recommend making this pythonic:

• Don’t use mathy variable names or dict parameters. Use expressive names in lower case, especially for the 3rd parameter onwards:

  • X → counts (In scanpy we do rows=observations, cols=features)
  • L → n_components
  • ctl["maxIter"] → iter_max
  • init["factors"] → init_factors
  • ...

  def glm_pca(counts, n_components: int, *, fam="poi", iter_max=1000, ...):

• Best would be to to use the scipy way of creating a class that takes all the model parameters and fitting it to the data, but keeping it one function is also fair.

• If you don’t use a class, Best don’t return a dict, but a namedtuple. This has the advantage that elements are easier to access, and it can be unpacked:

  # I just want the fist two items, don’t care about the rest:
  factors, loadings, *_ = glm_pca(counts, 50)
  
  # I want an object containing everything:
  pca = glm_pca(counts, 2, fam="muli")
  plt.scatter(*pca.factors.T)

[willtownes]

these are all good suggestions- thanks!