theislab / augurpy Goto Github PK

From the R augur implementation:
Execute a permutation test to identify cell types with statistically
significant differences in AUC between two different rounds of cell type
prioritization (for instance, the response to drugs A and B, as compared
to a common untreated control).

Future type annotations only work with Python 3.10+

• Document the API functions (copy structure from ehrapy @Zethson )
• Write an example notebook and add it to a tutorials folder. Ideally show it in the documentation as well. @antschum
• Document the technical differences and differences in result compared to the original R implementation @antschum
• Add top level module docstrings (currently missing, see below) @antschum
  

Implement plots visualising the feature importance.

eg. Barplot showing distribution of feature importances.

Think we should make many of the dependencies more loose and work with >= instead of '^='.
But we can do that in a follow up PR. Or I can do that :)

Originally posted by @Zethson in #32 (comment)

(Not sure if issues are intended to be used this way, but.)

I would like to add the feature importances for the random forest and logistic regression. In the case of the random forest I propose to use the build in function. For logistic regression there is no äquivalent on sklearn. Augur used standardized coefficients with Agresti method and linked this: https://think-lab.github.io/d/205/#3.

I have not had time to fully dive into it but was wondering if @Zethson you know any other way or have preferences :)
(This is not integral to creating a cell type ranking so its not super urgent.)