ttest_analysis <- function( df,
new.data.window = NULL,
min.days=7,
min.data.points=30,
test.type,
min.duration) {
df <- data.frame(df)
new.record.failed <- NULL
new.record <- NULL
df <- df[!is.na(df$duration),]
if (is.null(df)) return (NULL)
#if (nrow(df)<20) return (NULL)
if (nrow(df)<min.data.points |
length(unique(as.POSIXlt(df$timestamp))) < min.days)
return (NULL)
# sort out file dash and widget names
dash.id <- grep("dashboard",names(df))
dash.name <- substr(df[1,dash.id],start=0,40)
widg.id <- grep("widget",names(df))
widg.name <- as.character(df[1,widg.id])
version <- unique(df$sisenseVersion)[[1]]
cube.id <- grep("cubeName",names(df))
cube.name <- as.character(df[1,cube.id])
key.id <- grep("key",names(df))
key.name <- df[1,key.id]
user <- c(unique(df$sisenseuid))[[1]]
print(paste("dashboard : ", dash.name,sep=""))
# handle division of data to historical and new
if (new.data.window > 0)
{
# write logic for a predefined new data time window
df$timestamp <- as.POSIXlt(df$timestamp)
cutoff.date <- unique(max(df$timestamp)) - new.data.window
baseline.df <- subset(df, timestamp <= cutoff.date)
newdates.df <- subset(df, timestamp > cutoff.date)
if (nrow(newdates.df) < 1.5* new.data.window |
nrow(baseline.df) < 3 * new.data.window) return (NULL)
minDate <- min(as.Date (newdates.df$timestamp))
maxDate <- max(as.Date (newdates.df$timestamp))
minDate.baseline <- min(as.Date (baseline.df$timestamp))
maxDate.baseline <- max(as.Date (baseline.df$timestamp))
} else{
df.sorted <- df[order(as.Date(df$timestamp)),]
num.row <- nrow(df.sorted)
baseline.id <- ceiling(num.row*0.8)
baseline.df <- as.data.frame(df.sorted[1:baseline.id,c("duration","timestamp")]); names(baseline.df)[1] <- "duration"
newdates.df <- as.data.frame(df.sorted[(baseline.id+1):num.row,c("duration","timestamp","status")]);names(newdates.df)[1] <- "duration"
minDate <- min(as.Date (newdates.df$timestamp))
maxDate <- max(as.Date (newdates.df$timestamp))
minDate.baseline <- min(as.Date (baseline.df$timestamp))
maxDate.baseline <- max(as.Date (baseline.df$timestamp))
#version <- unique(newdates.df$sisenseVersion)[[1]]
if (is.null(version)) version = ""
}
if (any(newdates.df$status == "failed")) {
new.record.failed <- c(key.name,
paste("from ",minDate," to ",maxDate,sep=""),
"Query Failed",
"",
user,
version,
"",
"",
"",
"",
"",
"",
"",
"")
}
if (nrow(newdates.df) < 3 |
length(nearZeroVar(newdates.df$duration)) >0 |
length(nearZeroVar(baseline.df$duration))> 0 )
{
return (NULL)
}
# visualize baseline distribution
# calculate median and variance
med.base <- mean(baseline.df$duration)
med.new <- mean(newdates.df$duration)
if (med.new < min.duration)
return (NULL)
sd.base <- sd(baseline.df$duration)
sd.new <- sd(newdates.df$duration)
##############################################################
# t test
if (test.type == "parametric"){
p.value <- t.test(baseline.df$duration,
newdates.df$duration,
alternative = "less",
var.equal=FALSE,
paired=T)$p.value
significance.result <- p.value < 0.01
} else { # non-parametric (wilcox test)
p.value <- wilcox.test(baseline.df$duration,
newdates.df$duration,
alternative = "less",
paired=T)$p.value
#paired=F)$p.value
significance.result <- p.value < 0.05
}
# save to file
setwd("/home/sisense/Documents/anomaly/widget_outputs")
#fname <- substr(df$X56e932620ab266301900093c.Transaction.Dashboard..V2.2..dashboard[1],start=0,25)
file.name <- str_replace_all(dash.name, "[^[:alnum:]]", " ")
# writing results
if (significance.result) {
new.record <- c(key.name,
paste("from ",minDate," to ",maxDate,sep=""),
"Query_Duration_Deterioration",
p.value,
user,
version,
as.character(minDate.baseline),
as.character(maxDate.baseline),
as.character(minDate),
as.character(maxDate),
med.base,
sd.base,
med.new,
sd.new)
}
#t.test(baseline.df$duration,newdates.df$duration)$p.value)
#alerts.df <<- rbind(alerts.df, new.record)
if (!is.null(new.record.failed) & !is.null(new.record))
{
return(list(new.record,new.record.failed))
} else if (!is.null(new.record.failed) & is.null(new.record))
{
return(new.record.failed)
} else if (is.null(new.record.failed) & !is.null(new.record))
{
return(new.record)
}
}
