cupyx.scipy.sparse.bmat default return format is different than scipy.sparse.bmat

[ivirshup]

Description

Cupy's sparse bmat returns a coo matrix when calling bmat on a grid of csr matrices where scipy returns a csr matrix.

I would expect the result format to be the same.

To Reproduce

import cupyx.scipy.sparse as cpsparse
from scipy import sparse

X_gpu = cpsparse.random(20, 10, format="csr")
X_cpu = X_gpu.get()

assert sparse.bmat([[X_cpu, X_cpu], [X_cpu, X_cpu]]).format == "csr"

assert cpsparse.bmat([[X_gpu, X_gpu], [X_gpu, X_gpu]]).format == "csr"

---------------------------------------------------------------------------
AssertionError                            Traceback (most recent call last)
Cell In[4], [line 9](vscode-notebook-cell:?execution_count=4&line=9)
      [5](vscode-notebook-cell:?execution_count=4&line=5) X_cpu = X_gpu.get()
      [7](vscode-notebook-cell:?execution_count=4&line=7) assert sparse.bmat([[X_cpu, X_cpu], [X_cpu, X_cpu]]).format == "csr"
----> [9](vscode-notebook-cell:?execution_count=4&line=9) assert cpsparse.bmat([[X_gpu, X_gpu], [X_gpu, X_gpu]]).format == "csr"

AssertionError:

Installation

None

Environment

 OS                           : Linux-5.15.0-101-generic-x86_64-with-glibc2.35
Python Version               : 3.12.2
CuPy Version                 : 13.1.0
CuPy Platform                : NVIDIA CUDA
NumPy Version                : 1.26.4
SciPy Version                : 1.12.0
Cython Build Version         : 0.29.37
Cython Runtime Version       : None
CUDA Root                    : /usr/local/cuda
nvcc PATH                    : /usr/local/cuda/bin/nvcc
CUDA Build Version           : 12000
CUDA Driver Version          : 12040
CUDA Runtime Version         : 12000 (linked to CuPy) / 12040 (locally installed)
cuBLAS Version               : (available)
cuFFT Version                : 11201
cuRAND Version               : 10305
cuSOLVER Version             : (11, 6, 1)
cuSPARSE Version             : (available)
NVRTC Version                : (12, 4)
Thrust Version               : 200001
CUB Build Version            : 200200
Jitify Build Version         : <unknown>
cuDNN Build Version          : None
cuDNN Version                : None
NCCL Build Version           : None
NCCL Runtime Version         : None
cuTENSOR Version             : None
cuSPARSELt Build Version     : None
Device 0 Name                : Tesla T4
Device 0 Compute Capability  : 75
Device 0 PCI Bus ID          : 0000:00:06.0

Additional Information

There is similar behaviour with hstack:

assert sparse.hstack([X_cpu, X_cpu]).format == "csr"
assert not cpsparse.hstack([X_gpu, X_gpu]).format == "csr"

[takagi]

Thanks for your report! I could reproduce the your result, and it seems that cupy.scipy.sparse haven't been following scipy.sparse's changes for recent years.

Are you interested in contributing to cupy's bmat to make it get along with the latest scipy?