1703-06337.R

#
# 1703-06337.R, 28 Dec 19
# Data from:
# Mefta Sadat and Ayse Basar Bener and Andriy V. Miranskyy
# Rediscovery Datasets: Connecting Duplicate Reports
#
# Example from:
# Evidence-based Software Engineering: based on the publicly available data
# Derek M. Jones
#
# TAG fault-report_duplicate KDE_fault-report


source("ESEUR_config.r")


library("gnm")
library("plyr")

# 
# 
# plot_dups=function(df)
# {
# t=count(count(df$root_id)$freq)
# if (nrow(t) < 15)
#    return()
# 
# lines(t$x, t$freq)
# }


# plot_layout(2, 1)

pal_col=rainbow(4)

kde=read.csv(paste0(ESEUR_dir, "reliability/kde.csv.xz"), as.is=TRUE)


dups=subset(kde, !is.na(root_id))

# plot(0.9, type="n", log="xy",
# 	xlim=c(1, 20), ylim=c(1, 400),
# 	xlab="Reports", ylab="Duplicates")
# 
# d_ply(t, .(product), plot_dups)
# 
# 

dup_cnt=count(count(dups$root_id)$freq)

# plot(dup_cnt$freq, log="y", col=point_col,
#         xlab="Duplicates", ylab="Reports\n")
# 
# fail_mod=gnm(freq ~ instances(Mult(1, Exp(x)), 2)-1,
#                 data=dup_cnt[-1, ], verbose=TRUE, trace=TRUE,
#                 start=c(30000.0, -0.7, 300.0, -0.1),
#                 family=poisson(link="identity"))
# summary(fail_mod)
# 
# exp_coef=as.numeric(coef(fail_mod))
# 
# lines(exp_coef[1]*exp(exp_coef[2]*dup_cnt$x), col=pal_col[1])
# lines(exp_coef[3]*exp(exp_coef[4]*dup_cnt$x), col=pal_col[3])
# 
# t=predict(fail_mod)
# lines(t, col=pal_col[2])


plot(dup_cnt$x, dup_cnt$freq, log="y", col=point_col,
	xaxs="i",
	xlim=c(1, 170),
        xlab="Fault report ID", ylab="Occurrences\n")

# Cannot get any convergence if the first count is included
dup2_cnt=dup_cnt[-1, ]

fail_mod=gnm(freq ~ instances(Mult(1, Exp(x)), 3)-1,
                data=dup2_cnt, verbose=FALSE, trace=FALSE,
                start=c(230000.0, -1.0, 2100.0, -0.3, 21, -0.03),
                family=poisson(link="identity"))
summary(fail_mod)

exp_coef=as.numeric(coef(fail_mod))

lines(dup2_cnt$x, exp_coef[1]*exp(exp_coef[2]*dup2_cnt$x), col=pal_col[1])
lines(dup2_cnt$x, exp_coef[3]*exp(exp_coef[4]*dup2_cnt$x), col=pal_col[3])
lines(dup2_cnt$x, exp_coef[5]*exp(exp_coef[6]*dup2_cnt$x), col=pal_col[4])

t=predict(fail_mod)
lines(dup2_cnt$x, t, col=pal_col[2])


# library("pracma")
#
# # A different fit, constant offset makes a difference
# me_mod=mexpfit(dup2_cnt$x, dup2_cnt$freq, p0=c(-0.9, -0.1, -0.01))
# print(me_mod)