(
ar, return_index=False, return_inverse=False, return_counts=False
)
| 1708 | |
| 1709 | |
| 1710 | def unique_no_structured_arr( |
| 1711 | ar, return_index=False, return_inverse=False, return_counts=False |
| 1712 | ): |
| 1713 | # A simplified version of `unique`, that allows computing unique for array |
| 1714 | # types that don't support structured arrays (such as cupy.ndarray), but |
| 1715 | # can only compute values at the moment. |
| 1716 | |
| 1717 | if ( |
| 1718 | return_index is not False |
| 1719 | or return_inverse is not False |
| 1720 | or return_counts is not False |
| 1721 | ): |
| 1722 | raise ValueError( |
| 1723 | "dask.array.unique does not support `return_index`, `return_inverse` " |
| 1724 | "or `return_counts` with array types that don't support structured " |
| 1725 | "arrays." |
| 1726 | ) |
| 1727 | |
| 1728 | ar = ar.ravel() |
| 1729 | |
| 1730 | args = [ar, "i"] |
| 1731 | meta = meta_from_array(ar) |
| 1732 | |
| 1733 | out = blockwise(np.unique, "i", *args, meta=meta) |
| 1734 | out._chunks = tuple((np.nan,) * len(c) for c in out.chunks) |
| 1735 | |
| 1736 | out_parts = [out] |
| 1737 | |
| 1738 | name = f"unique-aggregate-{out.name}" |
| 1739 | dsk = { |
| 1740 | (name, 0): ( |
| 1741 | (np.unique,) |
| 1742 | + tuple( |
| 1743 | ( |
| 1744 | (np.concatenate, o.__dask_keys__()) |
| 1745 | if hasattr(o, "__dask_keys__") |
| 1746 | else o |
| 1747 | ) |
| 1748 | for o in out_parts |
| 1749 | ) |
| 1750 | ) |
| 1751 | } |
| 1752 | |
| 1753 | dependencies = [o for o in out_parts if hasattr(o, "__dask_keys__")] |
| 1754 | graph = HighLevelGraph.from_collections(name, dsk, dependencies=dependencies) |
| 1755 | chunks = ((np.nan,),) |
| 1756 | out = Array(graph, name, chunks, meta=meta) |
| 1757 | |
| 1758 | result = [out] |
| 1759 | |
| 1760 | if len(result) == 1: |
| 1761 | result = result[0] |
| 1762 | else: |
| 1763 | result = tuple(result) |
| 1764 | |
| 1765 | return result |
| 1766 | |
| 1767 |
no test coverage detected
searching dependent graphs…