Jesse Grabowski

`statespace` Roadmap: Enhancements and known bugs

2

A to-do list of things that still need to be done on the statespace models, to organize future PRs. Anyone is free to add to (or subtract from!) the list....

Implement `pt.pad`

2

## Description Implement `pt.pad`, following the `np.pad` API with feature parity. Very preliminary draft, uploading it in this state so I can ask @ricardoV94 to look at the `_linear_ramp_pad` function...

Add `einsum`

30

## Description TODO: - [x] Support ellipsis (related to https://github.com/dgasmith/opt_einsum/issues/235) - [x] Exclude broadcastable dims for better perf (JAX does the same) - [x] Handle missing static shape information (default...

### Description [Numpy](https://numpy.org/doc/stable/reference/generated/numpy.pad.html), along with derivative tensor libraries like [jax](https://jax.readthedocs.io/en/latest/_autosummary/jax.numpy.pad.html), [torch](https://pytorch.org/docs/2.3/generated/torch.nn.functional.pad.html), etc, all offer padding functions to insert values to the left and right (and top and bottom, etc) to...

Add rewrite to merge multiple `SVD` `Op`s with different settings

1

### Description SVD comes with a bunch of keyword arguments, most important of which is `compute_uv`. If False, it will return only the singular values for a given matrix. This...

Implement destructive in-place rewrites for Cholesky and Solve Ops

13

## Description All scipy.linalg functions offer an `overwrite_a` and/or `overwrite_b` argument that can enhance performance by re-using the input memory for the outputs. This PR implements re-writes that will set...

ENH: Implement COLA library rewrites for linear algebra functions

25

### Description [This paper ](https://arxiv.org/abs/2309.03060) and [this library ](https://github.com/wilson-labs/cola)describe and implement a number of linear algebra simplifications that we can implement as graph rewrites. This issue is both a tracker...

Add support for multiple output functions to `gradient.verify_grad`

1

### Description This is marked as a [to-do](https://github.com/pymc-devs/pytensor/blob/28f264834244723ae2ac098f79e139450d16cb36/pytensor/gradient.py#L1791) in the code, along with a suggestion of how to handle it. The whole function could probably be cleaned up while we're...

### Description `CholeskySolve` currently raises a non-implemented error on `L_op`, but it could probably just use the generic solve gradient defined in the `SolveBase` from which it inherits. In addition,...

Implement `lu_factor` and `lu_solve` for Blockwise `Solve` rewrites

1

### Description [This blog post ](https://www.johndcook.com/blog/2010/01/19/dont-invert-that-matrix/ )makes an important point about speed in cases where we want to repeatedly compute $A^{-1}b$ -- we should factorize the `A` matrix once, then...