aesara

aesara copied to clipboard

Closes #1094 - [ ] Inspect the types of graphs produced by the broadcasting used in this approach and make sure they're fully reduced. - [ ] Investigate the effects...

brandonwillard

Add docstring for `(Half)CauchyRV`, `InvGammaRV` and `WaldRV`

1

Also, for some reason, the `__call__` method of `InvGammaRV` required a `rate` parameter; I checked the scipy documentation, compared their definition to Wikipedia's and the distribution is indeed parametrized in...

rlouf

Fix cache versioning of empty `Op.__props__`

1

This PR closes #1101 by fixing the way C cache version tuples are generated when `Op.__props__` is empty. More specifically, these changes prevent the addition of empty tuples to the...

brandonwillard

Add Owens-T Function `Op`

5

This draft PR adds `scipy.special's` `owens_t`-function, needed primarily in the computation of a skewed Gaussian's CDF [1]. The gradient is taken from wolfram alpha ([2] and [3]). As I've never...

RuneDominik

Cache miss occurs due to use of hash() in key

3

## Description of your problem or feature request In certain circumstances we end up having incorrect cache misses leading to extra storage / CPU usage. To reproduce, save the following...

mattearllongshot

Missing `c_code_cache_version` in `grad(sum(a), a)` graph

2

## Description of your problem or feature request With this graph we do a compilation for each process that runs (storing it for the duration of the process), rather than...

mattearllongshot

Generalize `linalg.det` beyond 2D arrays

12

```python import numpy as np import aesara.tensor as at x = np.full((2, 3, 3), np.eye(3)) np.linalg.det(x) # broadcast operation fine at.linalg.det(x) # AssertionError x.ndim == 2 ```

purna135

Here's a MWE: ```python import aesara import aesara.tensor as at from aesara.raise_op import assert_op x = at.vector("x") x_a = assert_op(x, at.gt(x.shape[0], 1)) y = x_a.shape[0] aesara.dprint(y, print_type=True) # Subtensor{int64} [id...

brandonwillard

[`Elemwise` appears to store its `ufunc` objects in `Apply.tag`](https://github.com/aesara-devs/aesara/blob/fd50f36b404d1c89a841907d3682537d5f7719ab/aesara/tensor/elemwise.py#L673) in certain cases, and we absolutely do not want this—especially since this poor choice has already caused its concerns to leak...

brandonwillard

Use static shape in outputs of `Elemwise` and `Alloc` `Op`s

10

This PR allows for known shapes to be propagated to the outputs of Elemwise OPs It won't solve #732, but it doesn't seem like there was a good reason not...

ricardoV94