Numerical Utilities

EEGPrep includes the small numerical building blocks used by its signal and component workflows. They are standalone NumPy/SciPy functions and do not need MATLAB or an EEGLAB checkout.

Indexing and compatibility

Array indices are zero-based. This applies to peak frames returned by abspeak, assignments returned by hungarian and matcorr, permutation vectors, matsel selections, and shuffle axes. Kernel peak coordinates remain one-based because they describe the sampled coordinate system in the EEGLAB formulas rather than Python array indices.

covary deliberately retains EEGLAB’s grand-mean centering before computing column second moments; use NumPy’s variance functions when ordinary per-column centering is intended. means uses the number of finite observations when computing standard errors and leaves single-observation sample variances undefined. This fixes the historical EEGLAB helper’s underestimated standard errors when a group contains missing values. nan_std uses a centered calculation so large constant offsets do not erase small physiological variation through catastrophic cancellation.

quantile retains the midpoint empirical-probability rule used by the current EEGLAB tests. Integer counts, axes, permutations, and selections must be exact integers; fractional values are rejected instead of silently selecting the wrong samples. Functions that have a meaningful complex-valued contract, including PCA, interpolation, peak phase, and NaN-aware means, preserve complex data. Real-only functions reject complex input instead of discarding its imaginary component.

vectdata supports linear, cubic, and nearest-neighbor interpolation. MATLAB’s legacy biharmonic griddata v4 mode has no well-defined one-dimensional SciPy equivalent and raises NotImplementedError rather than substituting a different interpolator.

Peaks, summaries, and transforms

eegprep.abspeak(data[, frames_per_epoch])	Return absolute peaks, zero-based frames, and signs for each epoch.
eegprep.averef(data[, weights, sphere, ...])	Average-reference channel-major data and optionally transform ICA weights.
eegprep.covary(data)	Return EEGLAB's globally centered, unbiased column second moment.
eegprep.datlim(data)	Return the minimum and maximum of a nonempty numeric array.
eegprep.eucl(coordinates[, other])	Compute pairwise Euclidean distances between rows of point arrays.
eegprep.means(data[, groups])	Return per-group means, standard errors, variances, and group IDs.
eegprep.nan_mean(data[, axis])	Return means while ignoring NaNs along the selected dimension.
eegprep.nan_std(data[, axis])	Return sample standard deviations while ignoring NaNs.
eegprep.quantile(data, probabilities[, axis])	Return quantiles using MATLAB's midpoint empirical-probability rule.
eegprep.vectdata(data, times, *, timesout[, ...])	Interpolate along the final data axis and optionally smooth first.

Kernels

eegprep.gauss(frames, standard_deviations)	Return an EEGLAB-compatible Gaussian window with a unit peak.
eegprep.gauss2d(rows, columns[, sigma_rows, ...])	Return an EEGLAB-compatible two-dimensional Gaussian kernel.
eegprep.gauss3d(rows, columns, depth[, ...])	Return an EEGLAB-compatible three-dimensional Gaussian kernel.
eegprep.gabor2d(rows, columns[, frequency, ...])	Return a sinusoidal carrier under an anisotropic Gaussian envelope.
eegprep.laplac2d(rows, columns[, sigma, ...])	Return EEGLAB's sampled two-dimensional Laplacian kernel.

Matching and component projections

eegprep.hungarian(cost_matrix)	Return a zero-based column-to-row assignment and its total cost.
eegprep.mapcorr(first, second, ...[, ...])	Match rows after aligning map columns by common channel labels.
eegprep.matcorr(first, second[, ...])	Match rows of two matrices by absolute cosine correlation.
eegprep.matperm(first, second, ...)	Reorder and sign-correct rows of first into second row order.
eegprep.pcsquash(data[, components])	Compress channel-major data into its leading principal components.
eegprep.pcexpand(projections, eigenvectors, ...)	Expand component projections back into channel space.
eegprep.perminv(permutation)	Return the inverse of a zero-based permutation vector.
eegprep.uniquef(groups[, sort])	Return finite unique values, counts, and zero-based first indices.

Low-level compatibility helpers

eegprep.celltomat(cells)	Convert a rectangular nested sequence of numeric scalars to an array.
eegprep.eyelike(matrix)	Permute and scale rows so the output diagonal is one.
eegprep.fastif(condition, if_true, if_false)	Return if_true when condition is truthy, otherwise if_false.
eegprep.matsel(data, frames_per_epoch, frames)	Select zero-based channels, within-epoch frames, and epochs.
eegprep.mattocell(matrix)	Convert a NumPy-compatible matrix to nested Python scalar lists.
eegprep.scanfold(folder[, ignore, max_depth])	Return MATLAB filenames below folder and EEGLAB's -a text.
eegprep.shuffle(data[, axis, rng])	Shuffle one zero-based axis and return permutation and inverse indices.