meddlr.ops

Meddlr provides some utilities for complex, categorical, and fft-related operations. All operations assume the first dimension is the batch dimension.

Complex Utilities

These utilities are used to perform complex arithmetic with both complex tensors and real views of compex tensors, where the last dimension is 2 for real/imaginary components. Because these utilities are optimized for complex numbers, they will fail on real-valued tensors that do not follow the real-view convention described above.

Note, because these utilites support real-view tensors, if the last dimension is 2, it will be interpreted as a real-view of a complex tensor. In your code, we recommend using complex tensors when handling complex numbers to avoid ambiguity between real and complex numbers.

meddlr.ops.complex.abs	Computes the absolute value (magnitude) of a complex-valued input tensor (x).
meddlr.ops.complex.angle	Computes the phase of a complex-valued input tensor (x).
meddlr.ops.complex.center_crop	Apply a center crop to the input image or batch of complex images.
meddlr.ops.complex.channels_first	Permute complex-valued x to channels-first convention.
meddlr.ops.complex.channels_last	Permute complex-valued x to channels-last convention.
meddlr.ops.complex.conj	Computes the complex conjugate of complex-valued input tensor (x).
meddlr.ops.complex.from_polar	Computes real and imaginary values from polar representation.
meddlr.ops.complex.get_mask	Returns a binary mask for where x is nonzero with eps tolerance.
meddlr.ops.complex.imag	Gets imaginary component of complex tensor.
meddlr.ops.complex.is_complex	Returns if x is a complex-tensor.
meddlr.ops.complex.is_complex_as_real	Returns True if the tensor follows the real-view convention for complex numbers.
meddlr.ops.complex.matmul	Computes complex-valued matrix product of X and Y.
meddlr.ops.complex.mul	Multiplies two complex-valued tensors x and y.
meddlr.ops.complex.power_method	Iteratively computes first singular value of X using power method.
meddlr.ops.complex.real	Gets real component of complex tensor.
meddlr.ops.complex.rss	Compute the root-sum-of-squares (RSS) for complex inputs.
meddlr.ops.complex.svd	Computes singular value decomposition of batch of complex-valued matrices.
meddlr.ops.complex.to_numpy	Convert real-valued PyTorch tensor to complex-valued numpy array.
meddlr.ops.complex.to_tensor	Convert complex-valued numpy array to real-valued PyTorch tensor.

Categorical Utilities

These utilities are designed to perform categorical operations on tensors, such as converting between categorical and one-hot representations and converting from logits to probabilities.

meddlr.ops.categorical_to_one_hot	Converts categorical predictions to one-hot encoded predictions.
meddlr.ops.logits_to_prob	Converts logits to probabilities.
meddlr.ops.one_hot_to_categorical	Converts one-hot encoded predictions to categorical predictions.
meddlr.ops.pred_to_categorical	Converts one-hot encoded predictions or logits to category.

Fourier Transform Utilities

meddlr.ops.fft2c	Apply centered 2 dimensional Fast Fourier Transform.
meddlr.ops.fft3c	Apply centered 2 dimensional Fast Fourier Transform.
meddlr.ops.fftc	Apply 1D centered Fast Fourier Transform (FFT).
meddlr.ops.fftnc	Apply nD centered fast fourier transform.
meddlr.ops.fftshift	Similar to np.fft.fftshift but applies to PyTorch Tensors
meddlr.ops.ifft2c	Apply centered 2-dimensional Inverse Fast Fourier Transform.
meddlr.ops.ifft3c	Apply centered 2-dimensional Inverse Fast Fourier Transform.
meddlr.ops.ifftc	Apply centered 2 dimensional Fast Fourier Transform.
meddlr.ops.ifftnc	Apply nD centered inverse fast fourier transform.
meddlr.ops.ifftshift	Similar to np.fft.ifftshift but applies to PyTorch Tensors