Ricardo Vieira
Ricardo Vieira
I see, we always always get tricked by coords without values (as in we tend to forget they exist all the time). In that case the volatility check should be...
It seems to fail at something very basing, trying to call `jax.numpy.array(data, dtype=dtype)` on a numpy array
Perhaps try with `pytensor.config.floatX="float32"` before defining the model, maybe it has zero support for float64
> > It seems to fail at something very basing, trying to call `jax.numpy.array(data, dtype=dtype)` on a numpy array > > Doesn't work for `float64`, `jax.numpy.array(x, dtype="float64")` Yes that's the...
You're now getting errors deep inside numpyro, and have left PyMC/PyTensor. Could you try something simpler first? `model.compile_logp(mode="JAX")(model.initial_point())`?
@drbenvincent try something even simpler, a model just with `mu` and do `compile_logp(sum=False, mode="JAX")`
You need to eval it still with `(model.initial_point())` ```python import pytensor pytensor.config.floatX = "float32" import pymc as pm with pm.Model() as model: mu = pm.Normal("mu", 0, 1) model.compile_logp(sum=False, mode="JAX")(model.initial_point()) ```
Great. So the next question is, can you use numpyro to sample from that very simple model?
So it's still pretty broken
Sounds good and the perform method can even raise a nice error message if a graph tries to evaluate it