equal_frequency bins in highly_variable_genes

[davidhbrann]

Hi,

Using Seurat, in their variable gene function I've had some success using the equal_frequency option, where each bin contains an equal number of genes. Would it possible to implement this option in scanpy?

If you'd like I could submit a PR to implement this feature. I think it could be as simple as using pd.qcut instead of pd.cut or you could use a similar style as in the cell_ranger flavor with pd.cut(df['mean'], np.r_[-np.inf, np.percentile(df['mean'], np.arange(10, 105, 5)), np.inf]). I don't know how useful it would be, but I could also add the option to have more bins in the cell_ranger flavor by replacing np.arange(10,105,5) with np.linspace(10, 100, n_bins - 1).

Best,
David

[falexwolf]

These appear to be consistent, simple changes and I assume they would be covered by this test: https://github.com/theislab/scanpy/blob/master/scanpy/tests/test_highly_variable_genes.py

If you have any doubts about this, let's discuss before making a PR. Otherwise, I'm happy if you move forward with it.

[flying-sheep]

Pretty sure this is fixed now