sc.pp.scale and sc.pp.regress_out error on first run of copied object

[tessadgreen]

sc.pp.scale is giving me an error when I run it for the first time, then running fine on the second. This occurs when I run on an object generated using .copy()

In terms of the numbers at the end--I have 2176 cells and 1600 highly variable genes.

If I run on an object not generated using copy, I get "Trying to set attribute .obs of view, making a copy." but it finishes on first run.

If I try to regress out counts first, I get

One additional oddity--if I run sc.pphighly_variable_genes with flavor = 'seurat' instead of flavor = 'cell_ranger' and call sc.pp.regress_out(Bcell, 'n_counts') prior to running sc.pp.scale(Bcell, max_value = 10) I don't get any error. If I don't run regress_counts but have the 'seurat' flavor I get

Really not sure what's happening but figured I should let you know.

Thanks!

sc.settings.verbosity = 3

scanpy==1.4.3 anndata==0.6.21 umap==0.3.9 numpy==1.16.4 scipy==1.2.1 pandas==0.24.2 scikit-learn==0.21.2 statsmodels==0.9.0 python-igraph==0.7.1 louvain==0.6.1

[ivirshup]

Thanks for the bug report! I think we've just fixed the first and third issue in #729, but I'm not to sure about the second. Could you try updating to the newest release of AnnData and letting us know if the error still occurs?

Would you mind also letting us know if this error occurs when you use one of the built in datasets, like sc.datasets.pbmc3k()?

[tessadgreen]

Sorry for the delay on this! I upgraded to "scanpy==1.4.3+115.g1aecabf anndata==0.6.22.post1 umap==0.3.9 numpy==1.16.4 scipy==1.3.0 pandas==0.24.2 scikit-learn==0.21.2 statsmodels==0.10.0" and the issue is gone.

The pre-built dataset also works with the upgraded version.

[ivirshup]

Glad to hear it! We've just released v1.4.4 which has this fix in it, so you can use that if you don't want to be on the development branch.