jax-ml/jax · error · NotImplementedError

The 'out' argument to jnp.nanmean is not supported.

Error message

The 'out' argument to jnp.nanmean is not supported.

What it means

jnp.nanmean's 'out' argument exists only for NumPy signature compatibility. JAX is functional and cannot mutate a user-provided output buffer, so nanmean raises NotImplementedError when out is passed.

Solutions

  1. Drop out and capture the return value
  2. Rewrite buffer-reuse idioms as jitted functions; JIT already optimizes allocations

Example fix

// before
np.nanmean(x, out=result)
// after
result = jnp.nanmean(x)
Defensive patterns

Strategy: validation

Validate before calling

if out is not None:
    out = None  # or raise early with your own message
m = jnp.nanmean(a, axis=axis, dtype=dtype, where=where)

Prevention

When it happens

Trigger: Calling jnp.nanmean(a, out=buf) or via positional args jnp.nanmean(a, 0, None, True, buf-like).

Common situations: Porting NaN-ignoring statistics code from NumPy/SciPy that used out= for memory reuse; kwargs-forwarding wrappers.

Related errors


AI-assisted analysis of jax-ml/jax@1e1c6a8fc0 (2026-08-27). Data as JSON: /api/errors/bcab73e32f859cb4. Report an issue: GitHub.

Appendix: source

Thrown at jax/_src/numpy/reductions.py:1819

    ...                    [1, 1, 0, 1]], dtype=bool)
    >>> jnp.nanmean(x, axis=1, keepdims=True, where=where)
    Array([[ 3. ],
           [ 9. ],
           [-1.5]], dtype=float32)

    If ``where`` is ``False`` at all elements, ``jnp.nanmean`` returns ``nan``
    along the given axis.

    >>> where = jnp.array([[False],
    ...                    [False],
    ...                    [False]])
    >>> jnp.nanmean(x, axis=0, keepdims=True, where=where)
    Array([[nan, nan, nan, nan]], dtype=float32)
  """
  a = ensure_arraylike("nanmean", a)
  where = check_where("nanmean", where)
  if out is not None:
    raise NotImplementedError("The 'out' argument to jnp.nanmean is not supported.")
  if dtypes.issubdtype(a.dtype, np.bool_) or dtypes.issubdtype(a.dtype, np.integer):
    return mean(a, axis, dtype, out, keepdims, where=where)
  if dtype is None:
    dtype = dtypes.to_inexact_dtype(a.dtype)
  else:
    dtype = dtypes.check_and_canonicalize_user_dtype(dtype, "mean")
  nan_mask = lax.bitwise_not(lax._isnan(a))
  normalizer = sum(nan_mask, axis=axis, dtype=dtype, keepdims=keepdims, where=where)
  td = lax.div(nansum(a, axis, dtype=dtype, keepdims=keepdims, where=where), normalizer)
  return td


@export
@api.jit(static_argnames=('axis', 'dtype', 'keepdims'))
def nanvar(a: ArrayLike, axis: Axis = None, dtype: DTypeLike | None = None, out: None = None,
           ddof: int = 0, keepdims: bool = False,
           where: ArrayLike | None = None, mean: ArrayLike | None = None) -> Array:
  r"""Compute the variance of array elements along a given axis, ignoring NaNs.

View on GitHub (pinned to 1e1c6a8fc0)