Create empty DataFrame or Series which has correct dtype
(x, columns=None, index=None, meta=None)
| 22 | |
| 23 | |
| 24 | def _meta_from_array(x, columns=None, index=None, meta=None): |
| 25 | """Create empty DataFrame or Series which has correct dtype""" |
| 26 | |
| 27 | if x.ndim > 2: |
| 28 | raise ValueError( |
| 29 | "from_array does not input more than 2D array, got" |
| 30 | f" array with shape {x.shape!r}" |
| 31 | ) |
| 32 | |
| 33 | if index is not None: |
| 34 | from dask.dataframe import Index |
| 35 | |
| 36 | if not isinstance(index, Index): |
| 37 | raise ValueError("'index' must be an instance of dask.dataframe.Index") |
| 38 | index = index._meta |
| 39 | |
| 40 | if meta is None: |
| 41 | meta = meta_lib_from_array(x).DataFrame() |
| 42 | |
| 43 | if getattr(x.dtype, "names", None) is not None: |
| 44 | # record array has named columns |
| 45 | if columns is None: |
| 46 | columns = list(x.dtype.names) |
| 47 | elif np.isscalar(columns): |
| 48 | raise ValueError("For a struct dtype, columns must be a list.") |
| 49 | elif not all(i in x.dtype.names for i in columns): |
| 50 | extra = sorted(set(columns).difference(x.dtype.names)) |
| 51 | raise ValueError(f"dtype {x.dtype} doesn't have fields {extra}") |
| 52 | fields = x.dtype.fields |
| 53 | dtypes = [fields[n][0] if n in fields else "f8" for n in columns] |
| 54 | elif x.ndim == 1: |
| 55 | if np.isscalar(columns) or columns is None: |
| 56 | return meta._constructor_sliced( |
| 57 | [], name=columns, dtype=x.dtype, index=index |
| 58 | ) |
| 59 | elif len(columns) == 1: |
| 60 | return meta._constructor( |
| 61 | np.array([], dtype=x.dtype), columns=columns, index=index |
| 62 | ) |
| 63 | raise ValueError( |
| 64 | "For a 1d array, columns must be a scalar or single element list" |
| 65 | ) |
| 66 | else: |
| 67 | if np.isnan(x.shape[1]): |
| 68 | raise ValueError("Shape along axis 1 must be known") |
| 69 | if columns is None: |
| 70 | columns = list(range(x.shape[1])) if x.ndim == 2 else [0] |
| 71 | elif len(columns) != x.shape[1]: |
| 72 | raise ValueError( |
| 73 | "Number of column names must match width of the array. " |
| 74 | f"Got {len(columns)} names for {x.shape[1]} columns" |
| 75 | ) |
| 76 | dtypes = [x.dtype] * len(columns) |
| 77 | |
| 78 | data = {c: np.array([], dtype=dt) for (c, dt) in zip(columns, dtypes)} |
| 79 | return meta._constructor(data, columns=columns, index=index) |
| 80 | |
| 81 |