Unknown dtype dtype('int64') cannot be converted to ?gRMatrix.

[lazappi]

I'm having issues converting scipy matrices (possibly other things as well but this is the reproducible example).

import anndata2ri
from rpy2.robjects import globalenv
from rpy2.robjects.conversion import localconverter
from numpy.random import poisson
from scipy.sparse import csr_matrix

mat = csr_matrix(poisson(1, size=(100, 2000)))

with localconverter(anndata2ri.converter):
    globalenv["mat"] = mat

Traceback (most recent call last):
  File "/Users/luke.zappia/Documents/Code/theislab/extended-single-cell-best-practices/rpy2-test.py", line 10, in <module>
    globalenv["mat"] = mat
  File "/Users/luke.zappia/miniconda3/envs/interoperability/lib/python3.9/site-packages/rpy2/robjects/environments.py", line 35, in __setitem__
    robj = conversion.get_conversion().py2rpy(value)
  File "/Users/luke.zappia/miniconda3/envs/interoperability/lib/python3.9/functools.py", line 888, in wrapper
    return dispatch(args[0].__class__)(*args, **kw)
  File "/Users/luke.zappia/miniconda3/envs/interoperability/lib/python3.9/site-packages/rpy2/robjects/conversion.py", line 240, in _py2rpy
    raise NotImplementedError(
NotImplementedError: Conversion 'py2rpy' not defined for objects of type '<class 'scipy.sparse._csr.csr_matrix'>'

Conda environment (I have also tried older scipy versions with the same result)

channels:
  - conda-forge
  - bioconda
  - defaults
dependencies:
  - conda-forge::python=3.9.16
  - conda-forge::jupyterlab=3.6.3
  - conda-forge::scanpy=1.9.3
  - anndata2ri=1.2
  - bioconductor-basilisk=1.9.12
  - bioconductor-scuttle=1.8.0
  - bioconductor-singlecellexperiment=1.20.0
  - bioconductor-zellkonverter=1.8.0
  - ipywidgets=8.1.0
  - r-base=4.2
  - r-hdf5r=1.3.8
  - r-remotes=2.4.2
  - r-reticulate=1.30
  - r-sessioninfo=1.2.2
  - r-seurat=4.3.0
  - r-seuratobject=4.1.3
  - rpy2=3.5.11
  - session-info=1.0.0

[flying-sheep]

Hm, seems like there’s an issue with the docs, sorry! What happens if you use

with localconverter(anndata2ri._conv.full_converter()):
    globalenv["mat"] = mat

or

adata = AnnData(mat)

with localconverter(anndata2ri._conv.full_converter()):
    globalenv["adata"] = adata

[lazappi]

Initially I got:

ValueError: Unknown dtype dtype('int64') cannot be converted to ?gRMatrix.

But if I change to:

mat = csr_matrix(poisson(1, size=(100, 2000)), dtype=numpy.float32)

Then both of these work. Thanks!

Should this same code be used within a notebook or is it different in that case?

[flying-sheep]

In notebooks, the best version is to still use activate, but I’ll get around to fix #109 eventually.

rpy2 is getting rid of activate, that’s why it has to change.

I’ll leave this open to

• fix the float64 use case.
• make sure the valueerror contains info about the real issue (dtype)

[JonatanSahar]

This is still a problem. In my adata, there are certain arrays I'm unable to convert to float32, so I'm unable to bypass this issue..

[flying-sheep]

OK, I took the time to fix it. It uses base::to.numeric, which might or might not use R’s internal 64 bit integer type. 🤷