James Lucas

RESEARCH: Supporting column-major matrices

8

Gulp... Is this worth doing? It will allow us to integrate more nicely with BLAS/LAPACK - and we get things like free transposition as a result. We currently have a...

Overly strong bounds on singular matrix checks

1

In places we are checking if entries in the matrix are close to zero by using: `entry < n * T::min_positive_value()`, where n is a small integer. It is better...

SVD bugs

6

_Somewhat related to #17 and #46._ There are two known bugs in the SVD implementation. - [x] **When computing the bidiagonal blocks we use the wrong criteria.** ``` rust //...

The `iter_mut` method for `BaseMatrixMut` produces a `SliceIterMut`. For `Matrix` we can do better than this as we know that the matrix data is contiguous. This has notable effects on...

Some benchmarks and comparisons

1

It would be valuable to benchmark rulinalg. It would also be great to do comparisons against other libraries, both utilizing LAPACK and standalone - to see how well rulinalg holds...

It would be good to add a slicing equivalent for Vector (similar to `MatrixSlice` for `Matrix`). For `Vector` we can probably do this in a nicer way than `Matrix`. We...

Following requirements: - Should be behind a feature flag (to avoid new users running into dependency hell by default). - Hopefully will fail at compile time if bindings are not...

Implement Principal Component Analysis

5

[Rulinalg](https://github.com/AtheMathmo/rulinalg) now supports SVD - it would be nice to have PCA as a result.

The kernel based methods clone the training data

1

This is very space inefficient. Instead of doing this we could provide lifetimes and use references to the training data. Perhaps this should be tackled as part of a wider...

Improvements to linear models

7

The current techniques used in the linear models are sub-par and numerically unstable. There are better techniques available that we should use. This is most notably an issue for the...