Suppress NumbaWarning

[ricardoV94]

I get this all the time, not useful:

NumbaWarning: Cannot cache compiled function "numba_funcified_fgraph" as it uses dynamic globals (such as ctypes pointers and large global arrays)

[aseyboldt]

Yes, those are annoying. But maybe we should filter those out in pytensor itself?

[ricardoV94]

Yes, those are annoying. But maybe we should filter those out in pytensor itself?

Fair enough. I guess I reasoned here because nutpie uses the numba backend on behalf of users. However if the warning is always useless can just do it there.

[ricardoV94]

@aseyboldt any idea how to directly trigger the warning? I tried to define a function with a large array but that didn't do it.

import pytensor
import pytensor.tensor as pt

x = pt.vector("x", shape=(100_000,))
y = x + pt.random.normal(size=(10, 100_000)).eval()
fn = pytensor.function([x], y, mode="NUMBA")