Sakshi-2797

Scale

11

This contribution holds modifications to the existing 'scale function' which turns out to be a faster implementation than the original one. We have introduced flavors - default and use_fastpp ,...

Regress out

1

This contribution also holds a faster implementation for regress_out function in the form of a new function 'numpy_regress_out'. The numpy_regress_out modification provides an overall speedup of approx **~18x** in comparison...

Filter

1

In this contribution , we have introduced a new flow for filtering in the form of a function 'filter' in the already existing filtering implementation by combining 'filter cells' ,...

Parallel read

1

We have replaced the existing sc.read() implementation with our parallel read implementation using numba. By using this implementation , we get upto **31x faster** reading for large files as compared...

Nearest neighbours

1

We have replaced the existing nearest neighbor implementation from umap with scikit-learn's implementation. By using Intel extension for scikit-learn, this implementation can be upto 2x faster for large datasets.

Leiden, Louvain with Katana Graph

5

In this contribution , we have introduced a new flavour 'katana' in the already existing leiden and louvain implementation. The implementation of this new flavour is based on the KatanaGraph...

Changes in Layouts , Addition of new files

1

This contribution includes an alternate cpp implementation of the "optimize_layout_euclidean" function in layouts.py. The changes includes a faster cpp implementation of the optimization function by introducing separate files - umap_extend.cpp,...

Changes in Spectral Layout

2

This contribution includes a faster implementation of the spectral_layout function. The implementation replaces the eigenvalue compute with a faster MKL based method. The modification provides a speedup of **~9x** in...

The PR holds changes for creation of binaries for mm2-fast.