score_genes:n_bins can be off

[joshua-gould]

Ask for 25, get 26:

n_bins = 25
obs_avg = pd.Series(np.arange(100), index=np.arange(100))
n_items = int(np.round(len(obs_avg) / (n_bins - 1)))
obs_cut = obs_avg.rank(method='min') // n_items
len(obs_cut.unique())

Ask for 25, get 23:

n_bins = 25
values = np.arange(100)
values[0:10] = 0
values[11:20] = 1
obs_avg = pd.Series(values, index=np.arange(100))
n_items = int(np.round(len(obs_avg) / (n_bins - 1)))
obs_cut = obs_avg.rank(method='min') // n_items
len(obs_cut.unique())

[ivirshup]

What are the cases where this happens?

[joshua-gould]

See examples above. Thanks.

[ivirshup]

Could you provide an example of what’s going wrong? There’s no scanpy code in your example.

[joshua-gould]

The code I included in my example and below is taken directly from the scanpy.tl.score_genes source code:

n_items = int(np.round(len(obs_avg) / (n_bins - 1)))
obs_cut = obs_avg.rank(method='min') // n_items

[flying-sheep]

fixed in the next version!