jax.numpy.unique_values#

jax.numpy.unique_values(x, /, *, size=None, fill_value=None)[source]#

Return unique values from x, along with indices, inverse indices, and counts.

JAX implementation of numpy.unique_values(); this is equivalent to calling jax.numpy.unique() with equal_nan set to True.

Because the size of the output of unique_values is data-dependent, the function is not typically compatible with jit() and other JAX transformations. The JAX version adds the optional size argument which must be specified statically for jnp.unique to be used in such contexts.

Parameters:

  • x (ArrayLike) – N-dimensional array from which unique values will be extracted.

  • size (int | None | None) – if specified, return only the first size sorted unique elements. If there are fewer unique elements than size indicates, the return value will be padded with fill_value.

  • fill_value (ArrayLike | None | None) – when size is specified and there are fewer than the indicated number of elements, fill the remaining entries fill_value. Defaults to the minimum unique value.

Returns:

An array values of shape (n_unique,) containing the unique values from x.

Return type:

Array

See also

Examples

Here we compute the unique values in a 1D array:

>>> x = jnp.array([3, 4, 1, 3, 1])
>>> jnp.unique_values(x)
Array([1, 3, 4], dtype=int32)

For examples of the size and fill_value arguments, see jax.numpy.unique().