eegsmooth {eegkit}

Spatial and/or Temporal Smoothing of EEG Data

Description

Smooths single- or multi-channel electroencephalography (EEG) with respect to space and/or time. Uses the bigspline, bigtps, and bigssa functions (from bigsplines package) for smoothing.

Usage

eegsmooth(voltage, space = NULL, time = NULL, nknots = NULL,
          rparm = NULL, lambdas = NULL, skip.iter = TRUE,
          se.fit = FALSE, rseed = 1234)

Arguments

voltage

Vector of recorded EEG voltage at each row in space.

space

Matrix of electrode coordinates (in three-dimensions) at which EEG was recorded. If space=NULL, data are temporally smoothed only.

time

Vector of time points at which EEG was recorded. If time=NULL, data are spatially smoothed only.

nknots

Number of knots to sample for smoothing. Positive integer.

rparm

Rounding parameter(s) to use for smoothing. See Notes and Examples.

lambdas

Smoothing parameter(s) to use for smoothing.

skip.iter

If FALSE, iterative spatial-temporal smoothing is skipped. Ignored if space=NULL or time=NULL.

se.fit

If TRUE, standard errors of smoothed values are calculated.

rseed

Random seed to use for knot selection. Set rseed=NULL to obtain different knots each time, or set rseed to any positive integer to use a different random seed.

Value

For temporal smoothing only: an object of class "bigspline" (see bigspline).

For spatial smoothing only: an object of class "bigtps" (see bigtps).

For spatial-temporal smoothing: an object of class "bigssa" (see bigssa).

Note

For temporal smoothing only (i.e., space=NULL), the input rparm should be a positive scalar less than 1. Larger values produce faster (but less accurate) approximations. Default is 0.01, which I recommend for temporal smoothing; rparm=0.005 may be needed for particuarly rough signals, and rparm=0.02 could work for smoother signals.

For spatial smoothing only (i.e., time=NULL), the input rparm should be a positive scalar giving the rounding unit for the spatial coordinates. For example, rparm=0.1 rounds each coordinate to the nearest 0.1 (same as round(space,1)).

For spatial-temporal smoothing (i.e., both space and time are non-null), the input rparm should be a list of the form rparm=list(space=0.1,time=0.01), where the 0.1 and 0.01 can be replaced by your desired rounding parameters.

Setting rparm=NA will use the full data solution; this is more computationally expensive, and typically produces a solution very similar to using rparm=0.01 (see references).

Author(s)

Nathaniel E. Helwig <[e-mail]>

References

Helwig, N. E. (2013). Fast and stable smoothing spline analysis of variance models for large samples with applications to electroencephalography data analysis. Unpublished doctoral dissertation. University of Illinois at Urbana-Champaign.

Helwig, N.E. (2015). bigsplines: Smoothing Splines for Large Samples. http://CRAN.R-project.org/package=bigsplines

Helwig, N. E. & Ma, P. (2015). Fast and stable multiple smoothing parameter selection in smoothing spline analysis of variance models with large samples. Journal of Computational and Graphical Statistics, 24(3), 715-732.

Helwig, N. E. & Ma, P. (2016). Smoothing spline ANOVA for super large samples: Scalable computation via rounding parameters. Statistics and Its Interface, 9(4), 433-444.

Examples

##########   EXAMPLE 1: Temporal    ##########

# get "PZ" electrode of "c" subjects in "eegdata" data
data(eegdata)
idx <- which(eegdata$channel=="PZ" & eegdata$group=="c")
eegdata <- eegdata[idx,]

# temporal smoothing
eegmod <- eegsmooth(eegdata$voltage,time=eegdata$time)

# define data for prediction
time <- seq(min(eegdata$time),max(eegdata$time),length.out=100)
yhat <- predict(eegmod,newdata=time,se.fit=TRUE)

# plot results using eegtime
eegtime(time*1000/255,yhat$fit,voltageSE=yhat$se.fit,ylim=c(-4,4),main="Pz")



##########   EXAMPLE 2: Spatial   ##########

# get time point 65 (approx 250 ms) of "c" subjects in "eegdata" data
data(eegdata)
idx <- which(eegdata$time==65L & eegdata$group=="c")
eegdata <- eegdata[idx,]

# remove ears, nose, and reference (Cz)
idx <- c(which(eegdata$channel=="X"),which(eegdata$channel=="Y"),
         which(eegdata$channel=="nd"),which(eegdata$channel=="Cz"))
eegdata <- eegdata[-idx,]

# match to eeg coordinates
data(eegcoord)
cidx <- match(eegdata$channel,rownames(eegcoord))

# spatial smoothing
eegmod <- eegsmooth(eegdata$voltage,space=eegcoord[cidx,1:3])

# use dense cap for prediction
mycap <- levels(factor(eegdata$channel))
ix <- eegcapdense(mycap,type="2d",index=TRUE)
data(eegdense)
space <- eegdense[ix,1:3]
yhat <- predict(eegmod,newdata=space)

# plot results using eegspace
#eegspace(space,yhat)
eegspace(eegdense[ix,4:5],yhat)



##########   EXAMPLE 3: Spatial-Temporal (not run)   ##########

# # get "c" subjects of "eegdata" data
# data(eegdata)
# idx <- which(eegdata$group=="c")
# eegdata <- eegdata[idx,]

# # remove ears, nose, and reference (Cz)
# idx <- c(which(eegdata$channel=="X"),which(eegdata$channel=="Y"),
#          which(eegdata$channel=="nd"),which(eegdata$channel=="Cz"))
# eegdata <- eegdata[-idx,]

# # match to eeg coordinates
# data(eegcoord)
# cidx <- match(eegdata$channel,rownames(eegcoord))

# # spatial-temporal smoothing
# eegmod <- eegsmooth(eegdata$voltage,space=eegcoord[cidx,1:3],time=eegdata$time)

# # time main effect
# newdata <- list(time=seq(min(eegdata$time),max(eegdata$time),length.out=100))
# yhat <- predict(eegmod,newdata=newdata,se.fit=TRUE,include="time")
# eegtime(newdata$time,yhat$fit,voltageSE=yhat$se.fit,ylim=c(-2,4),main="Time Main Effect")

# # space main effect
# mycap <- levels(factor(eegdata$channel))
# ix <- eegcapdense(mycap,type="2d",index=TRUE)
# data(eegdense)
# newdata <- list(space=eegdense[ix,1:3])
# yhat <- predict(eegmod,newdata=newdata,include="space")
# eegspace(newdata$space,yhat)

# # interaction effect (spatial map at time point 65)
# newdata <- list(space=eegdense[ix,1:3],time=rep(65,nrow(eegdense[ix,])))
# yhat <- predict(eegmod,newdata=newdata,include="space:time")
# eegspace(newdata$space,yhat)

# # full prediction (spatial map at time point 65)
# newdata <- list(space=eegdense[ix,1:3],time=rep(65,nrow(eegdense[ix,])))
# yhat <- predict(eegmod,newdata=newdata)
# eegspace(newdata$space,yhat)