ARI_spots_cluster

[benklerc]

Changes proposed in this pull request:

• local-self-consistency method_get_ARI_spots_cluster added

• I am initializing the ARI_per_bin with 0, this leads to empty bins having a value of 0 in the returned ndarray, Is this alright?

• updated wrapper function self_consistency_metrics

• added 2 new parameter uns_key: where to find the spot data in adata_sp.uns ann_key: where to find the cell annotation in the adata_sp.uns[uns_key] I set the default values, as necessary for our simulated data, can of course be changed to more common ones

• do we need the cells_x_col/cells_y_col in self_consistency metircs?

• local cell_based method _get_number_of_celltypes added

• using _get_cell_density_grid_per_celltype

If new dependencies are needed:

sklearn.metrics necessary for ARI calculation

[LouisK92]

Great, just a little adjustment based on your question:

• empty bins should have the value np.nan, 0 would have the meaning that there's total dissimilarity
• keep cells_x_col/cells_y_col in the self_consistency metircs wrapper. The functions we currently want to support (ARI, and celltype_annotation_similarity) don't need it, but pretty sure there will be additional ones.

[benklerc]

fixed the empty bins to np.nan and added the most recent version of the get_number_of_cell_types with the 3D array and without the for-loop as discussed in the meeting