conda-forge/staged-recipes#9327

I've tried using this library and I have a few general comments that would, IMO, make this package much simple to use. To be clear, I was following the notebooks.

• You should add checks to make sure objects are of the correct type at the beginning of the script so it fails right away e.g. I passed in a sparse matrix instead of a dataframe and it was running for some time before it failed, and even then the message was quite cryptic.
• It's pretty strange you require the user to pass in a pandas dataframe, but then I thought about it a little and it makes sense since you need a way to match up the marker genes and the genes in the count matrix. Still, that caught me a little off guard.
• It seems strange to me that the annotations in the markers table need to have the name "Cell Type", after all, the library is called "point-annotator" and not "cell annotator". I think the most elegant solution would be to just allow a parameter annotation_col, which would allow me to quickly set and change which column I want to use for annotations.
• At first, I was completely confused by the num_genes parameter in assign_annotations. What does that have to do with anything? Reading the docstring did not help me out at all. Upon closer inspection, I realized that this is important for the DE test (I guess?). What would be the correct thing to do here if you don't know how many genes the organism has?
• You might want to add a function that does plotting with the annotations into the library itself. It's tedious to write code for plotting, and this is something most people would want to do (I guess).

Also, the notebooks are a little confusing, a lot of nontrivial code for plotting confusion matrices and all that. Other than that, it's really cool 😄