API / Docstrings

calculate_Σinv(Ξ;λ=0.)

Calculate the regularized inverse of the noise covariance matrix from the residuals Ξ, with optional regularization parameter λ. Regularization towards the diagonal is used, the Ledoit-Wolf shrinkage method.

Arguments

• Ξ: The residuals, used to estimate the noise covariance (ch x time)
• λ: Regularization parameter for the noise covariance estimation (default: 0.0 - recommended to be quite small in 2014 paper)

contained_or_equal(p, e)

Test if p equals e or whether e contains p if e is a tuple.

cvMANOVA(m,eeg)

Arguments

• m: A fitted UnfoldModel with a LinearModelFitCV modelfit.
• eeg: EEG data used to estimate the noise covariance matrix from the training data (chan x time x trials). Subset if you only want to use e.g. the baseline.

cvMANOVA(β_train,β_test,X_train,X_test,Y_train;C, C_test=C,λ=0.05)

A function to perform cross-validated MANOVA (cvMANOVA) on the training and test estimates from a cross validated modelfit This function calculates the cvMANOVA D statistic for each time point based on the training and test estimates, the design matrices. It estimates the noise covariance matrix estimated from the given training data.

Arguments

• β_train: The parameter estimates from the training data (chan, time, coeff).
• β_test: The parameter estimates from the test data (chan, time, coeff).
• X_train: The design matrix for the training data.
• X_test: The design matrix for the test data.
• Y_train: The observed data for the training set, used to estimate the noise covariance - subset if you only want to use e.g. the baseline.

Keyword Arguments

• C: The contrast matrix for the training data (no default).
• C_test: The contrast matrix for the test data (default: same as C).
• λ: Regularization parameter for the noise covariance estimation (default: 0.0 - recommended to be quite small in 2014 paper)

extract_coefs(model::UnfoldModel, predictor, basisname)

Return the coefficients of an Unfold model for a certain predictor and basisname.

For extracting the terms of a predictor variable predictor must be a symbol e.g. :continuous. For extracting the intercept predictor should be a String, i.e. "(Intercept)".

basisname must match one of the basis names which can be found in coeftable(model).

Note: If a predictor variable has more than one term in the formula (e.g. a spline set, a categorical variable with several levels or an interaction), the coefficients for all terms are returned.

The dimensions of the returned coefficients are channel x times x coefficients.

extract_coefs(models::Vector{<:UnfoldModel}, predictor, basisname)

When applied to a vector of Unfold models, extracts the coefficients (matching the predictor and basisname) for all models (usually subjects) and concatenates them.

The dimensions of the returned coefficients are channel x times x coefficients x subjects.

extract_symbol(t::AbstractTerm)

Return the symbol(s) underlying a term from a model formula, repeated by their actual coefficient number (after StatsModels.apply_schema).

Examples

julia> f = @formula 0 ~ 1 + spl(continuous, 4) + continuous + condition + pet + condition & pet
julia> ... # apply schema using an event dataframe, according to StatsModels
julia> extract_symbol(f)
8-element Vector{Any}:
 "(Intercept)"
 :continuous
 :continuous
 :continuous
 :continuous
 :condition
 :pet
 (:condition, :pet)

We get the actual symbols of each predictor - this is different to a function that would return the symbol for each term, which would be ["(Intercept)", :continuous,:continuous,:condition,:pet,(:condition,:pet) ]

The difference between those two cases would get even more stark, if a basisfunction is in play as it timeexpand terms into many more predictors.

get_predictor_string(p)

Return string representation based on the type of p.

This function is used for a useful display of variables e.g. in an error message.

Examples

julia> UnfoldStats.get_predictor_string(:condition)
":condition"