library(ggplot2)
library(reshape2)
library(glue)
library(dplyr)
#'Generate the 95% CI generated via bootstrapping
#
#'@param data the data to use during bootstrapping
#'@param statistic the statistical function to use
#'
#'@return an array containing (defaultStats, lowerBound, upperBound)
bootstrapCI_q = function(data, statistic)
{
b = boot(data, statistic=statistic, R=5000)
cis = boot.ci(b, conf=0.95, type="bca")
return(c(statistic(data), cis$bca[4], cis$bca[5]))
}
#'Generate the mean value of a list using the log anti log method (geometrical mean)
#'
#'@param x the original dataset
#'@param d the indice or list of indices to subsample 'x'
#'
#'@return exp(mean(log(x[d])))
geomMeanFunc = function(x, d)
{
return(exp(mean(log(x[d]))))
}
#'Plot a list of bootstrap data
#'
#'@param d the list of data.frame object with each containing: a column mean, lower, and upper. A name column permits to name the bar chart (each name is a metric). An id column allows to gather objects with different ids under the same metric.
#' You can change the column names with the corresponding label
#'@param maxAxis the maximum axis value
#'@param isBarChart is the graph a bar chart? (default: FALSE)
#'
#'@return the plot object
#'
#'
#'
#'
plotStackedBarchart = function(d, maxAxis=NA, minAxis=NA, legendName="",
xLabel=value, yLabel=name, fillLabel=id, facetLabel=grid)
{
xLabel = deparse(substitute(xLabel))
yLabel = deparse(substitute(yLabel))
fillLabel = deparse(substitute(fillLabel))
facetLabel = deparse(substitute(facetLabel))
d[[fillLabel]] = factor(d[[fillLabel]])
g = ggplot(d, aes(x = get(xLabel), y = get(yLabel), fill = factor(get(fillLabel), levels=unique(.data[[fillLabel]]), ordered=TRUE))) +
geom_bar(stat = 'identity', position = 'stack') + facet_grid(~ factor(get(facetLabel), levels=unique(.data[[facetLabel]]), ordered=TRUE))
if(is.na(maxAxis) && is.na(minAxis))
g = g + expand_limits(y = 0)
else
{
if(is.na(maxAxis))
maxAxis = 0
if(is.na(minAxis))
minAxis = 0
g = g + expand_limits(y = c(minAxis, maxAxis))
}
g = g + labs(x = "", y = "", col=legendName, fill=legendName) +
scale_y_continuous(labels = scales::percent_format()) +
scale_fill_brewer(palette="Paired") +
theme(plot.background = element_rect(fill = 'transparent', colour = 'transparent'),
panel.background = element_rect(fill = 'transparent', colour = 'transparent'),
plot.margin=grid::unit(c(0,0,0,0), "mm"),
axis.title = element_text(size = rel(1), colour = "black"),
axis.text = element_text(size = rel(1), colour = "black"),
panel.grid.major = element_line(colour = "#DDDDDD"),
panel.grid.major.y = element_blank(),
panel.grid.minor.y = element_blank(),
aspect.ratio = 3/4,
legend.title = element_text(size=12, face="bold"),
legend.text = element_text(size=10))
return(g)
}
plotListBootstrap = function(d, maxAxis=NA, minAxis=NA, isBarChart=FALSE, legendName="",
meanLabel=mean, lowerLabel=lower, upperLabel=upper,
nameLabel=name, idLabel=id)
{
lowerLabel = deparse(substitute(lowerLabel))
upperLabel = deparse(substitute(upperLabel))
meanLabel = deparse(substitute(meanLabel))
tr = data.frame(d)
tr = tr %>% group_by(!!ensym(idLabel)) %>% mutate(id = seq.int(1, nrow(cur_data()), 1))
# tr = tr[with(tr,order(-id)),] ## Sorting
tr$id = factor(tr$id, ordered=TRUE)
g = ggplot(tr, aes(fill=factor(.data[[substitute(idLabel)]], levels=unique(.data[[substitute(idLabel)]]), ordered=TRUE), x=id, y=get(meanLabel), colour=.data[[substitute(idLabel)]]))
if(isBarChart)
g = g + geom_bar(width=0.5, stat="identity", position=position_dodge())
if(is.na(maxAxis) && is.na(minAxis))
g = g + expand_limits(y = 0)
else
{
if(is.na(maxAxis))
maxAxis = 0
if(is.na(minAxis))
minAxis = 0
g = g + expand_limits(y = c(0, 6))
}
g = g +
scale_x_discrete(name="", labels=unique(tr[[substitute(nameLabel)]]), expand=c(0, 0)) +
scale_y_continuous(name="") +
geom_errorbar(aes(ymin=get(lowerLabel), ymax=get(upperLabel)), width = 0, size = 0.5, position=position_dodge(width=0.5), stat="identity") +
labs(x = "", y = "", col=legendName, fill=legendName) +
coord_flip() +
scale_color_brewer(palette="Paired") +
theme(plot.background = element_rect(fill = 'transparent', colour = 'transparent'),
panel.background = element_rect(fill = 'transparent', colour = 'transparent'),
plot.margin=grid::unit(c(0,0,0,0), "mm"),
axis.title = element_text(size = rel(1), colour = "black"),
axis.text = element_text(size = rel(1), colour = "black"),
panel.grid.major = element_line(colour = "#DDDDDD"),
panel.grid.major.y = element_blank(),
legend.title = element_text(size=13, face="bold"),
legend.text = element_text(size=10),
aspect.ratio = nrow(tr)/50) +
geom_point(size=1.5, position=position_dodge(width=0.5)) # dots
return(g)
}
meanFunc = function(x, d)
{
return(mean(x[d]))
}
# Parse the Questionnaire
parseQuestionnaire = function()
{
outputDir = "resultFiles/Questionnaire"
dir = "logfiles/Questionnaire"
data = NULL
#Read the csv files and concatenate them.
#They should have the same header
for(fileName in list.files(path=dir, pattern="*.csv"))
{
glue("Parsing {fileName}...")
#Parse
csvData = read.csv(glue("{dir}/{fileName}"), sep=',', header=TRUE)
#Concatenate
if(is.null(data))
data = csvData
else if(names(data) == names(csvData)) #This should be true
data = rbind(data, csvData)
else #If csv files are not consistent: exit the script
{
print("Error: csv files are not consistent. Exiting...")
stop()
}
}
print("Bootstraping TLX data for task 1-4")
tlxBootstrap1 = NULL
techIDNames = c("Point", "Brush", "Paint","Baseline") #ID as stored in the CSV
techLabelNames = c("Po", "Br", "Pa","Ba") #Label to use for rendering
for(i in 1:length(techIDNames))
{
techID = techIDNames[i]
techLabel = techLabelNames[i]
metrics = c("Performance_task1to4", "Mental_task1to4", "Physical_task1to4", "Temporal_task1to4", "Frustration_task1to4", "Effort_task1to4")
ylablenames= c("Performance", "Mental", "Physical", "Temporal", "Frustration", "Effort")
varNames = paste(techID, metrics, sep='')
for(j in 1:length(metrics))
{
var = varNames[j]
metric = metrics[j]
ylablename=ylablenames[j]
#Perform the bootstrap
print(glue("bootstraping var {var}"))
bsData = data.frame(t(c(bootstrapCI_q(data[[var]], meanFunc), var, ylablename, techLabel)))
print(bsData)
bsData[1:3] = as.double(bsData[1:3])
rownames(bsData) = c(var)
if(is.null(tlxBootstrap1))
tlxBootstrap1 = bsData
else
tlxBootstrap1 = rbind(tlxBootstrap1, bsData)
}
}
colnames(tlxBootstrap1) = c("mean", "lower", "upper", "varName", "metric", "tech")
dataNames = c("1", "2", "3", "4", "5")
dataLabels = c("Disk", "Rings", "Shell","Strings","Filaments")
print("Bootstraping TLX data for task 5")
tlxBootstrap2 = NULL
techIDNames = c("Point", "Brush", "Paint","Baseline") #ID as stored in the CSV
techLabelNames = c("Po", "Br", "Pa","Ba") #Label to use for rendering
for(i in 1:length(techIDNames))
{
techID = techIDNames[i]
techLabel = techLabelNames[i]
metrics = c("Performance_task5", "Mental_task5", "Physical_task5", "Temporal_task5", "Frustration_task5", "Effort_task5")
ylablenames= c("Performance", "Mental", "Physical", "Temporal", "Frustration", "Effort")
varNames = paste(techID, metrics, sep='')
for(j in 1:length(metrics))
{
var = varNames[j]
metric = metrics[j]
ylablename=ylablenames[j]
#Perform the bootstrap
print(glue("bootstraping var {var}"))
bsData = data.frame(t(c(bootstrapCI_q(data[[var]], meanFunc), var, ylablename, techLabel)))
print(bsData)
bsData[1:3] = as.double(bsData[1:3])
rownames(bsData) = c(var)
if(is.null(tlxBootstrap2))
tlxBootstrap2 = bsData
else
tlxBootstrap2 = rbind(tlxBootstrap2, bsData)
}
}
colnames(tlxBootstrap2) = c("mean", "lower", "upper", "varName", "metric", "tech")
dataNames = c("1", "2", "3", "4", "5")
dataLabels = c("Disk", "Rings", "Shell","Strings","Filaments")
print("Bootstraping Paired-Wise TLX data for task1 to 4")
pwTLXBootstrap1 = NULL
techIDPairedNames = data.frame(c("Brush", "Point"), #Name as stored in the CSV
c("Point", "Paint"),
c("Point", "Baseline"),
c("Brush", "Paint"),
c("Brush", "Baseline"),
c("Paint", "Baseline"))
techLabelPairedNames = data.frame(c("Br", "Po"), #Label to use for rendering
c("Po", "Pa"),
c("Po", "Ba"),
c("Br", "Pa"),
c("Br", "Ba"),
c("Pa", "Ba"))
for(i in 1:length(techIDPairedNames))
{
idPairedNames = t(techIDPairedNames[i])
labelPairedNames = t(techLabelPairedNames[i])
varNames = c("Performance_task1to4", "Mental_task1to4", "Physical_task1to4", "Temporal_task1to4", "Frustration_task1to4", "Effort_task1to4")
ylableNames = c("Performance", "Mental", "Physical", "Temporal", "Frustration", "Effort")
a=1
for(var in varNames)
{
print(glue("Bootstraping var {var}"))
tech1 = paste(idPairedNames[1], var, sep='')
tech2 = paste(idPairedNames[2], var, sep='')
d1 = data[[tech1]] + 1 #Discard infinite values
d2 = data[[tech2]] + 1 #Discard infinite values
ratio = d1/d2
ylableName=ylableNames[a]
bsData = data.frame(t(c(bootstrapCI_q(ratio, geomMeanFunc), ylableName, paste(labelPairedNames[1], labelPairedNames[2], sep=" / "))))
bsData[1:3] = as.double(bsData[1:3])
rownames(bsData) = paste(labelPairedNames[1], labelPairedNames[2], ylableName)
a=a+1
if(is.null(pwTLXBootstrap1))
pwTLXBootstrap1 = bsData
else
pwTLXBootstrap1 = rbind(pwTLXBootstrap1, bsData)
}
}
colnames(pwTLXBootstrap1) = c("mean", "lower", "upper", "varName", "tech")
print(pwTLXBootstrap1)
print("Bootstraping Paired-Wise TLX data for task5")
pwTLXBootstrap2 = NULL
techIDPairedNames = data.frame(c("Brush", "Point"), #Name as stored in the CSV
c("Point", "Paint"),
c("Point", "Baseline"),
c("Brush", "Paint"),
c("Brush", "Baseline"),
c("Paint", "Baseline"))
techLabelPairedNames = data.frame(c("Br", "Po"), #Label to use for rendering
c("Po", "Pa"),
c("Po", "Ba"),
c("Br", "Pa"),
c("Br", "Ba"),
c("Pa", "Ba"))
for(i in 1:length(techIDPairedNames))
{
idPairedNames = t(techIDPairedNames[i])
labelPairedNames = t(techLabelPairedNames[i])
varNames = c("Performance_task5", "Mental_task5", "Physical_task5", "Temporal_task5", "Frustration_task5", "Effort_task5")
ylableNames = c("Performance", "Mental", "Physical", "Temporal", "Frustration", "Effort")
a=1
for(var in varNames)
{
print(glue("Bootstraping var {var}"))
tech1 = paste(idPairedNames[1], var, sep='')
tech2 = paste(idPairedNames[2], var, sep='')
ylableName=ylableNames[a]
d1 = data[[tech1]] + 1 #Discard infinite values
d2 = data[[tech2]] + 1 #Discard infinite values
ratio = d1/d2
bsData = data.frame(t(c(bootstrapCI_q(ratio, geomMeanFunc), ylableName, paste(labelPairedNames[1], labelPairedNames[2], sep=" / "))))
bsData[1:3] = as.double(bsData[1:3])
rownames(bsData) = paste(labelPairedNames[1], labelPairedNames[2], ylableName)
a=a+1
if(is.null(pwTLXBootstrap2))
pwTLXBootstrap2 = bsData
else
pwTLXBootstrap2 = rbind(pwTLXBootstrap2, bsData)
}
}
colnames(pwTLXBootstrap2) = c("mean", "lower", "upper", "varName", "tech")
print(pwTLXBootstrap2)
print("Counting the TechniqueRank_dataset results...")
techIDNames = c("Point", "Brush", "Paint","BaseLine") #ID as stored in the CSV
techLabelNames = c("Po", "Br", "Pa","Ba") #Label to use for rendering
TechniqueRank_dataset = NULL
for(i in 1:length(techIDNames))
{
techName = techIDNames[i]
techLabel = techLabelNames[i]
print(glue("Counting for tech {techName}"))
#Names
Dataset1ColName = glue("D1{techName}Rank")
Dataset2ColName = glue("D2{techName}Rank")
Dataset3ColName = glue("D3{techName}Rank")
Dataset4ColName = glue("D4{techName}Rank")
Dataset5ColName = glue("D5{techName}Rank")
#Count
D1 = t(data.frame(c(sum(data[[Dataset1ColName]] == 1), "1st", "Disk", techLabel),
c(sum(data[[Dataset1ColName]] == 2), "2nd", "Disk", techLabel),
c(sum(data[[Dataset1ColName]] == 3), "3rd", "Disk", techLabel),
c(sum(data[[Dataset1ColName]] == 4), "4th", "Disk", techLabel)))
D2 = t(data.frame(c(sum(data[[Dataset2ColName]] == 1), "1st", "Rings", techLabel),
c(sum(data[[Dataset2ColName]] == 2), "2nd", "Rings", techLabel),
c(sum(data[[Dataset2ColName]] == 3), "3rd", "Rings", techLabel),
c(sum(data[[Dataset2ColName]] == 4), "4th", "Rings", techLabel)))
D3 = t(data.frame(c(sum(data[[Dataset3ColName]] == 1), "1st", "Shell", techLabel),
c(sum(data[[Dataset3ColName]] == 2), "2nd", "Shell", techLabel),
c(sum(data[[Dataset3ColName]] == 3), "3rd", "Shell", techLabel),
c(sum(data[[Dataset3ColName]] == 4), "4th", "Shell", techLabel)))
D4 = t(data.frame(c(sum(data[[Dataset4ColName]] == 1), "1st", "Strings", techLabel),
c(sum(data[[Dataset4ColName]] == 2), "2nd", "Strings", techLabel),
c(sum(data[[Dataset4ColName]] == 3), "3rd", "Strings", techLabel),
c(sum(data[[Dataset4ColName]] == 4), "4th", "Strings", techLabel)))
D5 = t(data.frame(c(sum(data[[Dataset5ColName]] == 1), "1st", "Filaments", techLabel),
c(sum(data[[Dataset5ColName]] == 2), "2nd", "Filaments", techLabel),
c(sum(data[[Dataset5ColName]] == 3), "3rd", "Filaments", techLabel),
c(sum(data[[Dataset5ColName]] == 4), "4th", "Filaments", techLabel)))
subTechniqueRank_dataset = rbind(D1, D2, D3,D4,D5)
TechniqueRank_dataset = rbind(TechniqueRank_dataset, subTechniqueRank_dataset)
}
TechniqueRank_dataset = data.frame(TechniqueRank_dataset)
colnames(TechniqueRank_dataset) = c("value", "TechniqueRank_dataset", "metric", "tech")
rownames(TechniqueRank_dataset) = NULL
TechniqueRank_dataset$value = as.numeric(TechniqueRank_dataset$value) / nrow(data)
print(TechniqueRank_dataset)
print("Counting the general ranking results...")
techIDNames = c("Point", "Brush", "Paint","BaseLine") #ID as stored in the CSV
techLabelNames = c("Po", "Br", "Pa","Ba") #Label to use for rendering
rank = NULL
for(i in 1:length(techIDNames))
{
techName = techIDNames[i]
techLabel = techLabelNames[i]
print(glue("Counting for tech {techName}"))
#Names
accColName = glue("acc{techName}Rank")
tctColName = glue("tct{techName}Rank")
genColName = glue("general{techName}Rank")
feaColName=glue("featureSelection{techName}Rank")
#Count
subAcc = t(data.frame(c(sum(data[[accColName]] == 1), "1st", "Accuracy", techLabel),
c(sum(data[[accColName]] == 2), "2nd", "Accuracy", techLabel),
c(sum(data[[accColName]] == 3), "3rd", "Accuracy", techLabel),
c(sum(data[[accColName]] == 4), "4th", "Accuracy", techLabel)))
subTct = t(data.frame(c(sum(data[[tctColName]] == 1), "1st", "Speed", techLabel),
c(sum(data[[tctColName]] == 2), "2nd", "Speed", techLabel),
c(sum(data[[tctColName]] == 3), "3rd", "Speed", techLabel),
c(sum(data[[tctColName]] == 4), "4th", "Speed", techLabel)))
subGen = t(data.frame(c(sum(data[[genColName]] == 1), "1st", "Overall", techLabel),
c(sum(data[[genColName]] == 2), "2nd", "Overall", techLabel),
c(sum(data[[genColName]] == 3), "3rd", "Overall", techLabel),
c(sum(data[[genColName]] == 4), "4th", "Overall", techLabel)))
subFea = t(data.frame(c(sum(data[[feaColName]] == 1), "1st", "Feature Selection", techLabel),
c(sum(data[[feaColName]] == 2), "2nd", "Feature Selection", techLabel),
c(sum(data[[feaColName]] == 3), "3rd", "Feature Selection", techLabel),
c(sum(data[[feaColName]] == 4), "4th", "Feature Selection", techLabel)))
subRank = rbind(subTct,subAcc ,subFea,subGen)
if(is.null(rank))
rank = subRank
else
rank = rbind(rank, subRank)
}
rank = data.frame(rank)
colnames(rank) = c("value", "rank", "metric", "tech")
rownames(rank) = NULL
rank$value = as.numeric(rank$value) / nrow(data)
print(rank)
#------------------------------------------------------------------------------
#---------------------------Plot questionnaire data----------------------------
#------------------------------------------------------------------------------
#Overall TLX
print(glue("Generating {outputDir}/tlx_task1to4.pdf"))
g = plotListBootstrap(tlxBootstrap1, maxAxis=2, nameLabel=metric, idLabel=tech)
ggsave(glue("{outputDir}/tlx_task1to4.pdf"), plot=g, device="pdf")
print(glue("Generating {outputDir}/tlx_task5.pdf"))
g = plotListBootstrap(tlxBootstrap2, maxAxis=2, nameLabel=metric, idLabel=tech)
ggsave(glue("{outputDir}/tlx_task5.pdf"), plot=g, device="pdf")
#PW comparisons
print(glue("Generating {outputDir}/PWtlx_task1to4.pdf"))
g = plotListBootstrap(pwTLXBootstrap1, nameLabel=varName, idLabel=tech) + geom_hline(yintercept = 1.0)
ggsave(glue("{outputDir}/PWtlx_task1to4.pdf"), plot=g, device="pdf")
print(glue("Generating {outputDir}/PWtlx_task5.pdf"))
g = plotListBootstrap(pwTLXBootstrap2, nameLabel=varName, idLabel=tech) + geom_hline(yintercept = 1.0)
ggsave(glue("{outputDir}/PWtlx_task5.pdf"), plot=g, device="pdf")
# #Counting technique rank by dataset
print(TechniqueRank_dataset)
print(glue("Generating {outputDir}/TechniqueRankbydataset.pdf"))
g = plotStackedBarchart(TechniqueRank_dataset, xLabel=tech, yLabel=value, fillLabel=TechniqueRank_dataset, facetLabel=metric)
ggsave(glue("{outputDir}/TechniqueRank_dataset.pdf"), plot=g, device="pdf")
#
#Counting Rank
print(rank)
print(glue("Generating {outputDir}/TechniqueRankGeneral.pdf"))
g = plotStackedBarchart(rank, xLabel=tech, yLabel=value, fillLabel=rank, facetLabel=metric)
ggsave(glue("{outputDir}/TechniqueRankGeneral.pdf"), plot=g, device="pdf")
}
parseQuestionnaire()
