Ralph Urlus

Replace Cython-based Arma -- Numpy conversions with Pybind11 + CARMA

12

### What is the desired addition or change? Current conversion from and to Numpy arrays is done with Cython which requires a copy both in and out on Windows. Armadillo...

Example/docs multiple windows

3

Hi Zachary, The project looks really good and working around a settable root solves my main frustration with tmuxinator. However, my setup is based on a session per project and...

Potentially mismatched allocator-deallocator when linking to external shared library

3

Found in #89, there is an issue with the free from `cnalloc.h` is used against a plain malloc used by armadillo when a library that includes carma is linked against...

Add pytest-leaks to CICD run

2

[Pytest-leaks](https://pypi.org/project/pytest-leaks/) could be an interesting addition to valgrind as it tracks the reference counts as well. It does require a debug-build of python.

Update of package & maintenance

1

Hi @alimuldal, Thank you for the package. I came across the repo and given the `no-maintenance intended` note updated it. Would you be willing to accept the PR with the...

[Feature Request] Interest in integrating uncertainty estimators for classification metrics?

2

Ha Team, great work on the library! We've worked on related functionality on the classification side in a package called [Model Metric Uncertainty (MMU)](https://github.com/RUrlus/ModelMetricUncertainty). The package was originally written as...

BUG: v1.13 introduced undocumented change in `sparse.csr_matrix` creation of matrices containing trailing column of all zeros

1

### Describe your issue. Release v1.13 caused a failure in our scheduled test-suite because it introduced a (in our case breaking) change in the instantiation of `csr_matrix` from a tuple...

Update sparse-dot-topn to v1

7

Hi @MaartenGr, We recently refactored [sparse-dot-topn](https://github.com/ing-bank/sparse_dot_topn) significantly and [released v1](https://github.com/ing-bank/sparse_dot_topn/releases/tag/v1.0.0). The most significant improvements are: - Faster implementation with lower memory overhead - new bindings using Nanobind which avoids the...

This PR implements the same strategy as used by scikit-learn to suppress the error (segfault) when OpenMP get loaded twice. This can happen when two or more packages vendor their...

`SparkCosSimIndexer. get_n_top_matches_for_all` has a type requirement for the `uid` column in that is must be convertible to a base 10 integer using the `int` constructor. This should probably be documented...