Transforms

GPU-accelerated transforms using PyTorch named tensors. Works on both CPU and GPU - tensors provide dimension awareness everywhere.

Base Classes

Transform

Temporal / Signal Processing

Temporal transforms for EMG feature extraction and signal processing.

SlidingWindowTransform(window_size[, ...])	Base class for sliding window transforms (GPU-accelerated).
RMS(window_size[, stride, dim])	Root Mean Square over sliding windows (GPU-accelerated).
MAV(window_size[, stride, dim])	Mean Absolute Value over sliding windows (GPU-accelerated).
VAR(window_size[, stride, dim])	Variance over sliding windows (GPU-accelerated).
Rectify([dim, name])	Full-wave rectification (absolute value).
Bandpass(low, high, fs[, order, Q, dim])	Bandpass filter using cascaded torchaudio biquads (GPU-accelerated).
Highpass(cutoff, fs[, order, Q, dim])	Highpass filter using torchaudio biquad (GPU-accelerated).
Lowpass(cutoff, fs[, order, Q, dim])	Lowpass filter using torchaudio biquad (GPU-accelerated).
Notch(freq[, width, fs, dim])	Notch filter using FFT (GPU-accelerated).
ZeroCrossings(window_size[, stride, dim])	Count zero crossings in sliding windows (GPU-accelerated).
SlopeSignChanges(window_size[, stride, dim])	Count slope sign changes in sliding windows (GPU-accelerated).
WaveformLength(window_size[, stride, dim])	Waveform length over sliding windows (GPU-accelerated).
Diff([n, dim])	Compute differences along a dimension.

Normalization

Normalization transforms for data preprocessing.

ZScore([dim, eps, keepdim])	Z-score normalization (mean=0, std=1) along a dimension.
MinMax([dim, eps, range])	Min-max normalization to [0, 1] range along a dimension.
Normalize([p, dim, eps])	L-p normalization along a dimension.
Standardize(mean, std[, eps])	Standardize using pre-computed mean and std.
InstanceNorm([eps])	Instance normalization (normalize each sample independently).
LayerNorm(normalized_shape[, eps])	Layer normalization along specified dimensions.
BatchNorm([eps])	Batch normalization (normalize over batch dimension).
ClampRange([min_val, max_val])	Clamp values to a specified range.

Generic Operations

Generic array operations.

Reshape(shape[, names])	Reshape tensor with new dimension names.
Index(indices[, dim])	Index/slice along a dimension.
Flatten([start_dim, end_dim])	Flatten dimensions of a tensor.
Squeeze([dim])	Remove dimensions of size 1.
Unsqueeze(dim[, name])	Add a dimension of size 1.
Transpose(dims, **kwargs)	Transpose/permute dimensions.
Mean([dim, keepdim])	Compute mean along a dimension.
Sum([dim, keepdim])	Compute sum along a dimension.
Stack(transforms[, dim])	Stack multiple tensors along a new dimension.
Concat(transforms[, dim])	Concatenate multiple tensors along an existing dimension.
Lambda(func, **kwargs)	Apply a custom function.
Identity([dim, name])	Identity transform (returns input unchanged).
Repeat(repeats[, dim])	Repeat tensor along a dimension.
Pad(padding[, dim, mode, value])	Pad tensor along a dimension.

Augmentation

Data augmentation transforms.

GaussianNoise([std, p])	Add Gaussian noise to the signal.
MagnitudeWarp([sigma, n_knots, p, dim])	Warp magnitude using smooth random curves.
TimeWarp([sigma, n_knots, p, dim])	Warp time axis with smooth random curves.
Dropout([p, dim])	Randomly zero out elements.
ChannelShuffle([p])	Randomly shuffle channel order.
TimeShift([max_shift, p, fill])	Randomly shift signal in time.
Scale([scale_range, p])	Random amplitude scaling.
Cutout([n_holes, length, p, dim])	Randomly zero out contiguous regions.

Spatial / Grid-Aware

Spatial transforms for electrode grid processing.

SpatialFilter([kernel, grids, dim])	Apply spatial filtering using grid layouts.
NDD([grids])	Normal Double Differential (Laplacian) filter.
LSD([grids])	Longitudinal Single Differential filter.
TSD([grids])	Transverse Single Differential filter.
IB2([grids])	Inverse Binomial 2nd order filter.