Jesse Grabowski

Isn't this exactly the jax graph but without the bisect logic to avoid checking every single value in `x` elemwise?

How do people do causality in STS framework? I'm interested in bringing more tools related to this into what we've already got in the statespace module. I know about granger...

Have you tried benchmarking the model with the JAX backend? I peeked at the code and saw it was an MvNormal likelihood. I see orders-of-magnitude speedups on statespace models (also...

The issue for this is https://github.com/pymc-devs/pymc/issues/7348 But the specific error @NathanielF reported [in his testing ](https://github.com/pymc-labs/CausalPy/pull/345#issuecomment-2155527736) is that there's no JAX funcify for LKJCholeskyCov, which needs another issue. TFP has...

Was there any solution to this issue? I am also getting these warnings when scanning over large tensors. I'd also be happy with some way to silence the warnings and...

I had in mind that it should be possible to the necessary replacements on the forward sampling graph directly, like model -> forward_graph -> frozen_forward_graph -> sampling_function, without making the...

I guess I had in mind that this hypothetical graph operation could just replace `freeze_rv_and_dims`. It's sort of a hack to pass go back to a model representation when we...

`freeze_dims_and_data` is only required for JAX, because it doesn't allow dynamic array shapes inside JIT functions. This isn't the case for numba jit functions (you can `@njit` a function then...

Also you can ignore the CI failure, it has nothing to do with your PR

So statsmodels was right of course. My equations above were slightly wrong. We want to have: $$ \begin{align} y_t &= \ell_{t-1} + \epsilon_t \\ \ell_t &= \ell_{t-1} + \alpha \epsilon_t...