enp.compat#

Compat utils between TF/Torch/Numpy/Jax.

Currently, each numpy API has slightly different behavior. Those functions ensure compatibility so that the code works seamlessly between all APIs.

In the future, those functions could be deleted and replaced by the official numpy API.

Symbols#


enp.compat.astype	`x.astype(dtype)`.
enp.compat.concat	`xnp.concatenate(x, axis=axis)`.
enp.compat.det	Like `np.linalg.det` but auto-support jnp, tnp, np.
enp.compat.dtype_np_to_torch	Returns the torch dtype for the given numpy dtype.
enp.compat.dtype_torch_to_np	Returns the numpy dtype for the given torch dtype.
enp.compat.expand_dims	`xnp.expand_dims(x, axis=axis)`.
enp.compat.inv	Like `np.linalg.inv` but auto-support jnp, tnp, np.
enp.compat.is_array_xnp	`isinstance(x, xnp.Array)`.
enp.compat.norm	Like `np.linalg.norm` but auto-support jnp, tnp, np.
enp.compat.round	`x.round()` for jnp, tnp, np, otrch.


enp.compat.lazy