Ricardo Vieira

Started failing after we bumped the dependency in #352 CC @ciguaran @aloctavodia

Several scipy special functions now upcast integers to float64

13

## Description Tests started failing due to Scipy 1.14.0 Unsure whether this will be a very prohibitive lower pin dependency. Clearly 0.14 was outdated (it's also poetic that the breaking...

Reductions along leading axes can be incredibly slow in C and Numba backends

2

### Description Reported by @aseyboldt ```python from timeit import timeit from functools import partial import pytensor import pytensor.tensor as pt import numpy as np from jax import block_until_ready N =...

Sketch of dim-ed tensors

1

```python import numpy as np from pytensor import function from pytensor.xtensor.basic import add, exp from pytensor.xtensor.type import xtensor x = xtensor("x", dims=("city",), shape=(None,)) y = xtensor("y", dims=("country",), shape=(4,)) z =...

Optimize `Sum`s of `MakeVector`s and `Join`s

3

### Please describe the purpose of filing this issue Not a drastic improvement by any means, but something we can keep in mind: ``` reduce(at.concatenate(*tensors)) -> reduce(reduce(tensor) for tensor in...

Implement proper JAX/Numba dispatch for IfElse

4

### Description IfElse is only lazy in the default backend because the function virtual machine handles it (via the "lazy" attribute"). In Numba/JAX it currently does nothing, because it receives...

### Description A matrix vector dot or a vector dot matrix can be represented as a matrix - col/row dot followed by a squeeze of the useless dims. I don't...

This rewrite only triggers for non-blockwise dots, but there's no reason not to also cover Blockwise dots with a Blockise Solve (if solve is not blockwised by default already) https://github.com/pymc-devs/pytensor/blob/ad27dc752798dfb75ed8d1c0a9b840859a8a898d/pytensor/tensor/rewriting/linalg.py#L90

Git rid of `ascontiguousarray` in numba reshape

1

## Description Seems not needed?

## Description Apparently both out implementation and numba's can be pretty dumb for reductions. This PR tries to default to the Numba case when possible. I am not sure this...