torchwrench.extras.numpy.scan_info module¶

class torchwrench.extras.numpy.scan_info.InvalidTorchDType(*args, **kwargs)[source]¶

Bases: object

Default return type for torch_dtype when an invalid data is passed as argument of scan_torch_dtype function. (like str for example)

class torchwrench.extras.numpy.scan_info.ShapeDTypeInfo(shape: tuple[int, ...], torch_dtype: torch.dtype | +T_Invalid | +T_EmptyTorch, numpy_dtype: numpy.dtype | +T_EmptyNp, valid_shape: bool)[source]¶

Bases: Generic[T_Invalid, T_EmptyTorch, T_EmptyNp]

property fill_value : bool | int | float | complex | None | str | bytes¶

property get_ndim : int¶

property kind : str¶

numpy_dtype : dtype | T_EmptyNp¶

shape : Tuple[int, ...]¶

torch_dtype : dtype | T_Invalid | T_EmptyTorch¶

valid_shape : bool¶

torchwrench.extras.numpy.scan_info.get_default_numpy_dtype() → dtype[source]¶

torchwrench.extras.numpy.scan_info.merge_numpy_dtypes(dtypes: Iterable[dtype | T_EmptyNp], *, empty: T_EmptyNp = dtype('V')) → dtype | T_EmptyNp[source]¶

torchwrench.extras.numpy.scan_info.merge_torch_dtypes(dtypes: Iterable[dtype | T_Invalid | T_EmptyNp], *, invalid: T_Invalid = InvalidTorchDType(), empty: T_EmptyNp = None) → dtype | T_Invalid | T_EmptyNp[source]¶

torchwrench.extras.numpy.scan_info.numpy_dtype_to_fill_value(dtype: Any) → bool | int | float | complex | None | str | bytes[source]¶

torchwrench.extras.numpy.scan_info.numpy_dtype_to_torch_dtype(dtype: dtype, *, invalid: T_Invalid = InvalidTorchDType()) → dtype | T_Invalid[source]¶

torchwrench.extras.numpy.scan_info.scan_numpy_dtype(x: Any, *, empty: T_EmptyNp = dtype('V')) → dtype | T_EmptyNp[source]¶

torchwrench.extras.numpy.scan_info.scan_shape_dtypes(x: Any, *, accept_heterogeneous_shape: bool = False, empty_torch: T_EmptyTorch = None, empty_np: T_EmptyNp = dtype('V')) → ShapeDTypeInfo[InvalidTorchDType, T_EmptyTorch, T_EmptyNp][source]¶

Returns the shape and the hdf_dtype for an input.

torchwrench.extras.numpy.scan_info.scan_torch_dtype(x: Any, *, invalid: T_Invalid = InvalidTorchDType(), empty: T_EmptyTorch = None) → dtype | T_Invalid | T_EmptyTorch[source]¶

Returns torch dtype of an arbitrary object. Works recursively on tuples and lists. An instance of InvalidTorchDType can be returned if a str is passed.

torchwrench.extras.numpy.scan_info.torch_dtype_to_numpy_dtype(dtype: dtype) → dtype[source]¶