Function quantile

dask/array/reductions.py:1565–1640

This works by automatically chunking the reduced axes to a single chunk if necessary and then calling ``numpy.quantile`` function across the remaining dimensions

(
    a,
    q,
    axis=None,
    out=None,
    overwrite_input=False,
    method="linear",
    keepdims=False,
    *,
    weights=None,
    interpolation=None,
)

@derived_from(np)
def quantile(
    a,
    q,
    axis=None,
    out=None,
    overwrite_input=False,
    method="linear",
    keepdims=False,
    *,
    weights=None,
    interpolation=None,
):
    """
    This works by automatically chunking the reduced axes to a single chunk if necessary
    and then calling ``numpy.quantile`` function across the remaining dimensions
    """
    if interpolation is not None:
        warnings.warn(
            "The `interpolation` argument to quantile was renamed to `method`.",
            FutureWarning,
            stacklevel=2,
        )

        if method != "linear":
            raise TypeError("Cannot pass interpolation and method keywords!")

        method = interpolation

    if axis is None:
        if builtins.any(n_blocks > 1 for n_blocks in a.numblocks):
            raise NotImplementedError(
                "The da.quantile function only works along an axis.  "
                "The full algorithm is difficult to do in parallel"
            )
        axis = tuple(range(len(a.shape)))
    elif not isinstance(axis, Iterable):
        axis = (axis,)
    axis = [ax + a.ndim if ax < 0 else ax for ax in axis]

    if overwrite_input:
        raise NotImplementedError("da.quantile does not support overwrite_input=True")

    # rechunk if reduced axes are not contained in a single chunk
    if builtins.any(a.numblocks[ax] > 1 for ax in axis):
        a = a.rechunk({ax: -1 if ax in axis else "auto" for ax in range(a.ndim)})

    q_arr = asarray_safe(q, like=a)
    kwargs = {}
    if weights is not None:
        if not NUMPY_GE_200:
            raise NotImplementedError("da.quantile with weights requires NumPy >=2.0")
        kwargs["weights"] = asarray_safe(weights, like=a)

    # weights break meta inference (dummy-array shape mismatches weights length).
    # There are also issues on NumPy 1.x.
    dtype = np.quantile(
        np.zeros(1, dtype=a.dtype),
        np.asarray(0.5, dtype=q_arr.dtype),