# -*- coding: utf-8 -*- import numbers import dask.array as da import numpy as np def _get_freq_grid(shape, chunks, axis, n, dtype=float): assert len(shape) == len(chunks) shape = tuple(shape) dtype = np.dtype(dtype).type assert (issubclass(dtype, numbers.Real) and not issubclass(dtype, numbers.Integral)) axis = axis % len(shape) freq_grid = [] for ax, (s, c) in enumerate(zip(shape, chunks)): if axis == ax and n > 0: f = da.fft.rfftfreq(n, chunks=c).astype(dtype) else: f = da.fft.fftfreq(s, chunks=c).astype(dtype) freq_grid.append(f) freq_grid = da.meshgrid(*freq_grid, indexing="ij", sparse=True) return freq_grid def _get_ang_freq_grid(shape, chunks, axis, n, dtype=float): dtype = np.dtype(dtype).type assert (issubclass(dtype, numbers.Real) and not issubclass(dtype, numbers.Integral)) pi = dtype(np.pi) freq_grid = _get_freq_grid(shape, chunks, axis, n, dtype=dtype) ang_freq_grid = tuple((2 * pi) * f for f in freq_grid) return ang_freq_grid def _norm_args(a, s, n=-1, axis=-1): if issubclass(a.dtype.type, numbers.Integral): a = a.astype(float) if isinstance(s, numbers.Number): s = np.array(a.ndim * [s]) elif not isinstance(s, da.Array): s = np.array(s) if issubclass(s.dtype.type, numbers.Integral): s = s.astype(a.real.dtype) elif not issubclass(s.dtype.type, numbers.Real): raise TypeError("The `s` must contain real value(s).") if s.shape != (a.ndim,): raise RuntimeError( "Shape of `s` must be 1-D and equal to the input's rank." ) if n != -1 and a.shape[axis] != (n // 2 + 1): raise NotImplementedError( "In the case of real-valued images, it is required that " "(n // 2 + 1) == image.shape[axis]." ) return (a, s, n, axis) def _reshape_nd(arr, ndim, axis): """Promote a 1d array to ndim with non-singleton size along axis.""" nd_shape = (1,) * axis + (arr.size,) + (1,) * (ndim - axis - 1) return arr.reshape(nd_shape)