Checking if an object is an array

[Jakob-Unfried]

Is there a best practice to verify that an object is an array, e.g. during input checks?
Comparing types seems unfeasable, since the function will be be called with abstract array types during tracing.

Essentially i want to do something like

import jax.numpy as np

def is_array(obj):
   return type(obj) == np.ndarray

But also consider all jax types that might occur during tracing.

If it doesn't already exist, such an is_array function would be nice to have, especially if the abstract types used during racing should change in the future

[jekbradbury]

I think isinstance(x, jax.numpy.ndarray) does what you want (it'll return True for JAX arrays, including abstract ones, as well as NumPy ndarrays and subtypes of ndarrays)

[mattjj]

@jekbradbury 's answer is the right one; though we might revise it (see #2014) right now you should check isinstance(x, jax.numpy.ndarray).

Hope that answers your question!