Brandon T. Willard
Brandon T. Willard
The scalar multiplication `Op`, `Mul`, has an `impl` method that actually uses `np.product`. When used in conjunction with `Elemwise`, optimizations like `local_mul_canonizer` construct graphs that essentially have `Elemwise(Mul)(a, b, c,...
The `tests.tensor.test_basic` module takes a while to run (e.g. another case of #23 ), and a big part of that is the unreasonably repetitious numeric testing that's performed in all...
The `deprecated` wrapper works fine for deprecating _functions_ but not for renamed classes and class instances—e.g. we don't want to break things by changing the expected type of an object....
Like Dask does with Pandas/NumPy (see [here](https://github.com/dask/dask/blob/1d0262c303b00fb053500d775ddb83437c6fcfe0/dask/utils.py#L666)), we can use a decorator that copies the docstrings from the NumPy functions that we're emulating/wrapping and adds a disclaimer at the end...
We might be able to speed up Aesara's compiled function evaluations by Cythonizing classes like [`Function`](https://github.com/pymc-devs/aesara/blob/84936418af87e84322bd668c7e05e62766699906/aesara/compile/function/types.py#L236)—or at least the methods within it and/or the functions it calls. Every time an...
We can't afford to write or maintain documentation for external libraries like NumPy or even Python itself—no matter how big or small the material may be. The two most egregious...
The `CAReduce` `Op` base class doesn't have a general gradient implementation, but it seems like it could.
We still don't have a gradient implementation for `Eig` (and a few other `aesara.linalg` `Op`s like https://github.com/aesara-devs/aesara/issues/836). [Here's](https://github.com/aesara-devs/aesara/pull/1020#issuecomment-1175405721) an example implementation that could work.
Some broken broadcasting-related logic is causing `local_elemwise_alloc` to produce graphs that fail to perform simple multiplications: ```python import numpy as np import aesara import aesara.tensor as at from aesara.compile.mode import...
This PR implements https://github.com/aesara-devs/aesara/issues/695. It's currently just an outline.