Add new unstack function to numpy/array_api namespaces

Author: Micky774

CHANGELOG.md

@@ -8,6 +8,10 @@ Remember to align the itemized text with the first line of an item within a list
 
 ## jax 0.4.27
 
+* New Functionality
+  * Added {func}`jax.numpy.unstack`, following the addition of this function in
+    the array API 2023 standard, soon to be adopted by NumPy.
+
 * Changes
   * {func}`jax.pure_callback` and {func}`jax.experimental.io_callback`
     now use {class}`jax.Array` instead of {class}`np.ndarray`. You can recover

jax/_src/numpy/lax_numpy.py

@@ -1887,6 +1887,12 @@ def stack(arrays: np.ndarray | Array | Sequence[ArrayLike],
       new_arrays.append(expand_dims(a, axis))
     return concatenate(new_arrays, axis=axis, dtype=dtype)
 
+@util.implements(getattr(np, 'unstack', None))
+@partial(jit, static_argnames="axis")
+def unstack(x: np.ndarray | Array, /, *, axis: int = 0) -> tuple[Array, ...]:
+  util.check_arraylike("unstack", x)
+  return tuple(moveaxis(x, axis, 0))
+
 @util.implements(np.tile)
 def tile(A: ArrayLike, reps: DimSize | Sequence[DimSize]) -> Array:
   util.check_arraylike("tile", A)

jax/experimental/array_api/__init__.py

@@ -173,6 +173,7 @@
     roll as roll,
     squeeze as squeeze,
     stack as stack,
+    unstack as unstack,
 )
 
 from jax.experimental.array_api._searching_functions import (

jax/experimental/array_api/_manipulation_functions.py

@@ -72,3 +72,7 @@ def stack(arrays: tuple[Array, ...] | list[Array], /, *, axis: int = 0) -> Array
   """Joins a sequence of arrays along a new axis."""
   dtype = _result_type(*arrays)
   return jax.numpy.stack(arrays, axis=axis, dtype=dtype)
+
+def unstack(x: Array, /, *, axis: int = 0) -> tuple[Array, ...]:
+  """Splits an array in a sequence of arrays along the given axis."""
+  return jax.numpy.unstack(x, axis=axis)

jax/numpy/__init__.py

@@ -253,6 +253,7 @@
     unpackbits as unpackbits,
     unravel_index as unravel_index,
     unsignedinteger as unsignedinteger,
+    unstack as unstack,
     unwrap as unwrap,
     vander as vander,
     vdot as vdot,

jax/numpy/__init__.pyi

@@ -859,6 +859,7 @@ def unpackbits(
 ) -> Array: ...
 def unravel_index(indices: ArrayLike, shape: Shape) -> tuple[Array, ...]: ...
 unsignedinteger = _np.unsignedinteger
+def unstack(x: _np.ndarray | Array, /, *, axis: int = ...) -> tuple[Array, ...]: ...
 def unwrap(p: ArrayLike, discont: Optional[ArrayLike] = ...,
            axis: int = ..., period: ArrayLike = ...) -> Array: ...
 def vander(

tests/array_api_test.py

@@ -163,6 +163,7 @@
   'unique_counts',
   'unique_inverse',
   'unique_values',
+  'unstack',
   'var',
   'vecdot',
   'where',

tests/lax_numpy_test.py

@@ -173,6 +173,20 @@ def f():
               for a in out]
     return f
 
+
+  @jtu.sample_product(
+    [dict(shape=shape, axis=axis)
+      for shape in all_shapes
+      for axis in list(range(-len(shape), len(shape)))],
+    dtype=all_dtypes,
+  )
+  def testUnstack(self, shape, axis, dtype):
+    rng = jtu.rand_default(self.rng())
+    x = rng(shape, dtype)
+    y = jnp.array(jnp.unstack(x, axis=axis))
+    self.assertArraysEqual(jnp.moveaxis(y, 0, axis), x)
+
+
   @parameterized.parameters(
     [dtype for dtype in [jnp.bool, jnp.uint8, jnp.uint16, jnp.uint32,
                          jnp.uint64, jnp.int8, jnp.int16, jnp.int32, jnp.int64,