scalableBayesEEG

R package for "Mohammed, S. and Dey, D.K., 2020. Scalable spatio-temporal Bayesian analysis of high-dimensional electroencephalography data. The Canadian Journal of Statistics, pp.1-20."

Code to run the model in (Mohammed et.al, 2020) on the EEG data using the package scalableBayesEEG.

Contents

1. Load EEG data and divide it into training and test.
2. Specify temporal and spatial covariance matrices
3. Build the model proposed in the manuscript with the training data and predict for the test data set.

# install the package (devtools package needed)
if(!require(devtools)) install.packages("devtools")
devtools::install_github('shariq-mohammed/scalableBayesEEG')

# Load the package
library(scalableBayesEEG)

Load EEG data and divide it into training and test

Load EEG Data

X = scalableBayesEEG::X_eeg

Load location indices for which distance matrix is available

ind64to57 = scalableBayesEEG::ind64to57_eeg

Update EEG data for locations above

X = X[,ind64to57,]

Load binary response

y = t(scalableBayesEEG::y_eeg)

Load distance matrix

dist.mat = scalableBayesEEG::dist.mat_eeg

Dimensions of EEG data

tau = dim(X)[1]
p = dim(X)[2]
n = dim(X)[3]

Sample indices for training data

trn.ind = sample.int(n, size = 100)

Responses for training and testing

y.trn = y[trn.ind] # responses from training data
y.test = y[-trn.ind] # responses from test data

EEG data for training and testing

X.trn = aperm(X[,,trn.ind], c(3,2,1)) # training EEG data
X.test = aperm(X[,,-trn.ind], c(3,2,1)) # test EEG data

Specify temporal and spatial covariance matrices

Specify temporal covariance matrix

prior.temp = matrix(0, nrow = tau, ncol = tau)
for(t1 in 1:tau) for(t2 in t1:tau){
  if(t1 == t2) prior.temp[t1,t1] = 2
  if(t2 == t1+1) prior.temp[t1,t2] = prior.temp[t2,t1] = -1
  if(t1 == 1 & t2 == 1) prior.temp[t1,t1] = 1
  if(t1 == tau & t2 == tau) prior.temp[t1,t1] = 1
}
T.mat = prior.temp+diag(0.6,tau)
Temp.cov = cov2cor(chol2inv(chol(T.mat))) # temporal covariance matrix

Specify spatial covariance matrix

dmat = (dist.mat+t(dist.mat))/2
W.mat = exp(-(dmat^2)/0.1)
D.mat = diag(rowSums(W.mat)+0.6)
G.mat = D.mat-W.mat
Spat.cov = cov2cor(chol2inv(chol(G.mat))) # spatial covariance matrix

Build the model proposed in the manuscript with the training data and predict for the test data set

Run Bayesian modeling with approximate estimation

finalModel = spTempBayes(y.trn, X.trn, c_init = rep(1, tau), beta_init = NULL,
                         zeta_init = NULL, nusq_init = NULL, l_init = NULL,
                         prior.mu = rep(0, p*tau), Spat.cov, Temp.cov,
                         v0 = 0.005, a1 = 5, a2 = 15, q = 1,
                         Nmcmc = 210, # takes about 120 seconds
                         burnin = 10, X.test = X.test, type='approx')

Plot heatmap of posterior estimate of the betas (rows are time points, columns are locations)

if(!require(lattice)) install.packages("lattice")
lattice::levelplot(finalModel$b.strucBayes)