#GAM Generalized Additive (Mixed) Models
#Week 12 slides

#in Praat, for each voice portion, file>extract selected sound (time from zero)
#rename each with the following pattern
# <participant initials lower case>_<gender>_<native/nonnative>_<t1/t2/t3/t4 (variation)>
#Select all sound objects (renamed ones only), and convert to pitch objects iusing analayse Periodicity > to pitch...
#convert > down to pitch tier


library(rPraat)
library(reshape2)
ptFiles <- dir(pattern = "*.PitchTier")
numFiles <- length(ptFiles)
numSamples <- 100

f0Contour <- matrix(numeric(numSamples*numFiles), ncol = numSamples)

for(fIdx in 1:numFiles)  {

  pitchData <- pt.read(ptFiles[fIdx])
  numTimeObs <- length(pitchData$t)
  timeVecInterp <- seq(pitchData$t[1], pitchData$t[numTimeObs], length = numSamples)
  f0Contour[fIdx, ] <- approx(pitchData$t, pitchData$f, timeVecInterp)$y
}

id <- as.factor(regmatches(ptFiles, regexpr("[a-z]*(?=\\_)", ptFiles, perl = TRUE)))
sex <- as.factor(regmatches(ptFiles, regexpr("[a-z]*(?=\\_)", ptFiles, perl = TRUE)))
native <- as.factor(regmatches(ptFiles, regexpr("[a-z]*(?=\\_)", ptFiles, perl = TRUE)))
tone <- as.factor(regmatches(ptFiles, regexpr("(?<=\\_)t.", ptFiles, perl = TRUE)))


dfMaster <- data.frame(id, sex, native, tone, f0Contour)
dfMelt <- melt(dfMaster, id.vars = c("id", "tone", "type", "tone"))

library(mgcv)
dfMelt$time <- as.numeric(dfMelt$variable)
fit <- bam(value ~ tone +
             s(time, by = tone, bs = "cr"), data = dfMelt, method = "fREML")
library(itsadug)
plot_smooth(fit, view= "time" , plot_all = c("tone") , rug = F)