Rémi Louf
Rémi Louf
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...
For API compatibility. `numpy.multiply` is an alias for *, the element-wise multiplication. https://numpy.org/doc/stable/reference/generated/numpy.multiply.html?highlight=multiply#numpy.multiply
The following code uses a `RandomStream` to define random variables and buid a simple logprob function with `joint_logprob`: ```python from aesara.graph.fg import FunctionGraph from aesara.link.numba.dispatch import numba_funcify import aesara.tensor as...
As in https://numpy.org/doc/stable/reference/generated/numpy.matmul.html
The JAX linker currently target the library's numpy-like high level API. For instance, the `Dot` Op is translated using `jax.numpy.dot`: ```python @jax_funcify.register(Dot) def jax_funcify_Dot(op, **kwargs): def dot(x, y): return jnp.dot(x,...
Evaluating etuplized objects fails for some `RandomVariable` ops whose `__call__` function does not defer to `make_node`. In this commit we wrap Ops during etuplization with a class that always defers...
`eval_if_etuple` does not work when naively evaluating etuplized random variables. Let us consider the following model: ```python import aesara.tensor as at from etuples import etuple, etuplize srng = at.random.RandomStream(0) Y_rv...
The following code fails with a `RecursionError`: ```python from kanren import eq, lall, lany, run, vars observations = [(0, 1), (1, 0)] def generate(pixels): pairs = [(a, b) for a,...
To use custom gradients users currently need to pass the function as a keyword arguments; this makes the API more complicated as we need to carry this argument downstream. We...
It's very simple: https://jax.readthedocs.io/en/latest/notebooks/Custom_derivative_rules_for_Python_code.html#python-debugging