$\%\%===$ constants $===$

$\%$ the time window relative to each event that may be used for classification

epoch$\_$range $=[-0\mathrm{.}20\mathrm{.}8]$;

$\%$ time ranges over which per$-$channel features should be extracted

time$\_$ranges $=[0\mathrm{.}250\mathrm{.}3$; $0\mathrm{.}30\mathrm{.}35$; $0\mathrm{.}350\mathrm{.}4$; $0\mathrm{.}40\mathrm{.}45$; $0\mathrm{.}450\mathrm{.}5$; $0\mathrm{.}50\mathrm{.}6]$;

$\%$ regularization parameter for the shrinkage LDA

lambda $=0\mathrm{.}1$;

$\%\%===$ train an ERP classifier $===$

$\%$ load the calibration data set

load ERP$\_$CALIB

$\%$ identify the sample latencies at which relevant events occur and their

$\%$ associated target class $(=0$ no error, $1=$error$)$

train$\_$events $=$ strcmp$(\{$ERP$\_$CALIB.event.type$\},\text{'}$S $11\text{'})|$ strcmp$(\{$ERP$\_$CALIB.event.type$\},\text{'}$S $12\text{'})|$ strcmp$(\{$ERP$\_$CALIB.event.type$\},\text{'}$S $13\text{'})$;

train$\_$latencies $=$ round$([$ERP$\_$CALIB.event$($train$\_$events$).$latency$])$;

train$\_$labels $=($~strcmp$(\{$ERP$\_$CALIB.event$($train$\_$events$).$type$\},\text{'}$S $11\text{'}))*2-1$;

model $=$ train$\_$erp$($ERP$\_$CALIB.data,ERP$\_$CALIB.srate,train$\_$latencies,train$\_$labels,epoch$\_$range,time$\_$ranges,lambda$)$;

$\%\%===$ apply the classifier to each event in the test data $===$

load ERP$\_$TEST

$\%$ determine the relevant event latencies and true labels

test$\_$events $=$ strcmp$(\{$ERP$\_$TEST.event.type$\},\text{'}$S $11\text{'})|$ strcmp$(\{$ERP$\_$TEST.event.type$\},\text{'}$S $12\text{'})|$ strcmp$(\{$ERP$\_$TEST.event.type$\},\text{'}$S $13\text{'})$;

test$\_$latencies $=$ round$([$ERP$\_$TEST.event$($test$\_$events$).$latency$])$;

test$\_$labels $=($~strcmp$(\{$ERP$\_$TEST.event$($test$\_$events$).$type$\},\text{'}$S $11\text{'}))*2-1$;

$\%$ also get the sample range that is used to extract epochs relative to the events

epoch$\_$samples $=$ round$($epoch$\_$range$(1)*$ERP$\_$TEST.srate$)$ : round$($epoch$\_$range$(2)*$ERP$\_$TEST.srate$)$;

$\%$ for each test event...

predictions $=[]$;

for e$=1$:length$($test$\_$latencies$)$

$\%$ extract the epoch

EPO $=$ ERP$\_$TEST.data$($:$,$epoch$\_$samples $+$ test$\_$latencies$($e$))$;

$\%$ classify it and record the prediction

predictions$($e$)=$ test$\_$erp$($EPO$,$model$)$;

end

$\%\%===$ evaluate the loss on the test set $===$

loss $=$ eval$\_$mcr$($test$\_$labels,predictions$)$;

fprintf$(\text{'}$The mis$-$classification rate on the test set is $\%\mathrm{.}2$f percent.

$\text{'},100*$loss$)$;