Bio-SeqAlignment-Examples-EnhancingEdlib
view release on metacpan or search on metacpan
scripts/plot_timings.R view on Meta::CPAN
library(data.table)
library(ggplot2)
library(mgcv)
library(viridis)
library(gratia)
library(patchwork)
rm(list = ls())
dat <- fread("timings.txt")
datOpenMP <- fread("timings_openMP.txt")
txtsize = 2
ggplot(dat, aes(
x = Workers,
y = Time,
color = factor(Chunk_Size)
)) + geom_point(size = 0.5) +
geom_line(linewidth = 0.4) +
scale_y_continuous(trans = 'log2', breaks = c(8, 16, 32, 64, 128, 256)) + theme_bw() + theme(
panel.grid.minor = element_blank(),
panel.grid.major = element_blank(),
axis.text.x = element_text(size = 8),
legend.position="bottom"
) + scale_color_viridis_d(name = "Chunk Size") +
scale_x_continuous(
trans = 'log2',
breaks = c(1, 2, 4, 8, 12, 18, 27, 36, 45, 54, 63, 72)
) +
geom_vline(
xintercept = c(18, 36, 54, 72),
linetype = "dashed",
color = "gray40"
) +
annotate(
"segment",
x = 36,
xend = 72,
y = 32,
yend = 32,
color = "gray80",
arrow = arrow(
type = "open",
ends = "both",
length = unit(0.1, "inches")
)
) +
geom_vline(
xintercept = c(18, 36, 54, 72),
linetype = "dashed",
color = "gray40"
) +
annotate(
"text",
x = 17,
y = 128,
label = "NUMA limit",
hjust = 1
) +
annotate(
"text",
x = 54,
y = 38,
label = "HYPERTHREADS",
hjust = 0.6,
lineheight = 0.5,
size = txtsize * 1.2
) +
annotate(
"text",
x = 8,
y = 70,
label = "node 0",
hjust = 0.5,
size = txtsize
) +
annotate(
"text",
x = 24,
y = 70,
label = "node 1",
hjust = 0.2,
size = txtsize
) +
scripts/plot_timings.R view on Meta::CPAN
label = "node 0",
hjust = 0.0,
size = txtsize
) +
annotate(
"text",
x = 56,
y = 70,
label = "node 1",
hjust = 0,
size = txtsize
)+
xlab("Number of Workers") + ylab("Time (seconds)")
ggsave(
"timings.png",
width = 6,
height = 4,
units = "in",
dpi = 1200
)
################################################################################
## combine openMP and MCE timings
dat1$Type <- "MCE"
datOpenMP$Type <- "OpenMP"
dat1$TotWorkers <- dat1$Workers
datOpenMP$TotWorkers <- datOpenMP$Workers * datOpenMP$Num_threads
datOMPCME <-
rbind(dat1[, c("Time", "Type", "TotWorkers", "Workers")],
datOpenMP[Workers == 1, c("Time", "Type", "TotWorkers", "Workers")])
datOMPCME$Type <-
factor(datOMPCME$Type, levels = c("MCE", "OpenMP"))
quantile(datOMPCME$TotWorkers, c(0.05,0.1, 0.9))
f2 <-
gam(log2(Time) ~ s(log2(TotWorkers), bs = "ts", by = Type) + Type, data = datOMPCME)
pr2 <- predict(f2, se = TRUE, type = "response")
datOMPCME$pr2 = 2 ^ pr2$fit
datOMPCME$prmin2 = 2 ^ (pr2$fit - 1.96 * pr2$se.fit)
datOMPCME$prmax2 = 2 ^ (pr2$fit + 1.96 * pr2$se.fit)
txtsize = 2
ggplot(datOMPCME, aes(x = TotWorkers, y = Time, shape = Type)) +
geom_point(size = 0.5,aes(color=Type)) +
geom_line(
data = datOMPCME,
aes(x = TotWorkers, y = pr2),
color = "black",
linewidth = 0.4
) +
geom_ribbon(data = datOMPCME,
aes(x = TotWorkers, ymin = prmin2, ymax = prmax2),
alpha = 0.2) +
theme_bw() + theme(
panel.grid.minor = element_blank(),
panel.grid.major = element_blank(),
axis.text.x = element_text(size = 8),
legend.position="bottom"
) +
scale_y_continuous(trans = 'log2', breaks = c(8, 16, 32, 64, 128, 256)) +
scale_x_continuous(
trans = 'log2',
breaks = c(1, 2, 4, 8, 12, 18, 27, 36, 45, 54, 63, 72)
) +
geom_vline(
xintercept = c(18, 36, 54, 72),
linetype = "dashed",
color = "gray40"
) +
annotate(
"segment",
x = 36,
xend = 72,
y = 32,
yend = 32,
color = "gray80",
arrow = arrow(
type = "open",
ends = "both",
length = unit(0.1, "inches")
)
) +
geom_vline(
xintercept = c(18, 36, 54, 72),
linetype = "dashed",
color = "gray40"
) +
annotate(
"text",
x = 17,
y = 128,
label = "NUMA limit",
hjust = 1
) +
annotate(
"text",
x = 54,
y = 38,
label = "HYPERTHREADS",
hjust = 0.6,
lineheight = 0.5,
size = txtsize * 1.2
) +
annotate(
"text",
x = 8,
y = 70,
label = "node 0",
hjust = 0.5,
size = txtsize
) +
annotate(
"text",
x = 24,
y = 70,
label = "node 1",
hjust = 0.2,
size = txtsize
) +
annotate(
"text",
x = 41,
y = 70,
label = "node 0",
hjust = 0.0,
size = txtsize
) +
annotate(
"text",
x = 56,
y = 70,
label = "node 1",
hjust = 0,
size = txtsize
)+
xlab("Number of Workers") + ylab("Time (seconds)")
ggsave(
"timings_MCE_OMP.png",
width = 6,
height = 4,
units = "in",
dpi = 1200
)
################################################################################
## now analyze the cost of memory for a given sec of performance
f3 <-
gam(log2(Workers * Time) ~ s(log2(TotWorkers), bs = "ts", by = Type) + Type, data = datOMPCME)
pr3 <- predict(f3, se = TRUE, type = "response")
datOMPCME$pr3 = 2 ^ pr3$fit
datOMPCME$prmin3 = 2 ^ (pr3$fit - 1.96 * pr3$se.fit)
datOMPCME$prmax3 = 2 ^ (pr3$fit + 1.96 * pr3$se.fit)
txtsize = 2
ggplot(datOMPCME, aes(x = TotWorkers, y = Workers * Time, shape = Type)) +
geom_point(size = 0.5,aes(color=Type)) +
geom_line(
data = datOMPCME,
aes(x = TotWorkers, y = pr3),
color = "black",
linewidth = 0.4
) +
geom_ribbon(data = datOMPCME,
aes(x = TotWorkers, ymin = prmin3, ymax = prmax3),
alpha = 0.2) +
theme_bw() + theme(
panel.grid.minor = element_blank(),
panel.grid.major = element_blank(),
axis.text.x = element_text(size = 8),
legend.position="bottom"
) +
scale_y_continuous(trans = 'log2',breaks =2^(4:9)) +
geom_segment(
x = log2(36),
xend = log2(72),
y = log2(32),
yend = log2(32),
color = "gray80",
arrow = arrow(
type = "open",
ends = "both",
length = unit(0.1, "inches")
)
) +
scale_x_continuous(
trans = 'log2',
breaks = c(1, 2, 4, 8, 12, 18, 27, 36, 45, 54, 63, 72)
) +
geom_vline(
xintercept = c(18, 36, 54, 72),
linetype = "dashed",
color = "gray40"
) +
annotate(
"segment",
x = 36,
xend = 72,
y = 32,
yend = 32,
color = "gray80",
arrow = arrow(
type = "open",
ends = "both",
length = unit(0.1, "inches")
)
) +
geom_vline(
xintercept = c(18, 36, 54, 72),
linetype = "dashed",
color = "gray40"
) +
annotate(
"text",
x = 17,
y = 128,
label = "NUMA limit",
hjust = 1
) +
annotate(
"text",
x = 54,
y = 38,
label = "HYPERTHREADS",
hjust = 0.6,
lineheight = 0.5,
size = txtsize * 1.2
) +
annotate(
"text",
x = 8,
scripts/plot_timings.R view on Meta::CPAN
annotate(
"text",
x = 24,
y = 70,
label = "node 1",
hjust = 0.2,
size = txtsize
) +
annotate(
"text",
x = 41,
y = 70,
label = "node 0",
hjust = 0.0,
size = txtsize
) +
annotate(
"text",
x = 56,
y = 70,
label = "node 1",
hjust = 0,
size = txtsize
)+
xlab("Number of Workers") + ylab("Resource Use ( cp x s)")
ggsave(
"spacetime_MCE_OMP.png",
width = 6,
height = 4,
units = "in",
dpi = 1200
)
################################################################################
datOpenMP$SpaceTime <- datOpenMP$Time * datOpenMP$Workers
## fit for time
f4<-gam(log2(Time)~ti(log2(Workers),bs="ts")+
ti(log2(Num_threads),bs="ts")+
ti(log2(Workers),log2(Num_threads),bs="ts"),data=datOpenMP)
datOpenMP$fitTime <- predict(f4,type="response")
## fit for space time
f5<-gam(log2(SpaceTime)~ti(log2(Workers),bs="ts")+
ti(log2(Num_threads),bs="ts")+
ti(log2(Workers),log2(Num_threads),bs="ts"),data=datOpenMP)
datOpenMP$fitSpaceTime <-predict(f5,type="response")
## find the 5% rows with the smallest Time & SpaceTime in the datOpenMP data
minTime<-datOpenMP[order(Time)][1:round(nrow(datOpenMP)*0.05)]
minSpaceTime<-datOpenMP[order(Time*Workers)][1:round(nrow(datOpenMP)*0.05)]
## three dimensional plots as contourplots
ggplot(datOpenMP, aes(x = Workers, y = Num_threads, z = fitTime)) +
geom_contour(aes(colour = after_stat(level))) +theme_bw() + theme(
panel.grid.minor = element_blank(),
panel.grid.major = element_blank(),
axis.text.x = element_text(size = 8),
legend.position="bottom"
) +
scale_color_viridis(discrete=FALSE,name="Time (log2 sec)") +
geom_point(data=minTime,aes(x=Workers,y=Num_threads),color="black",shape="+")+
scale_y_continuous(trans = 'log2', breaks = c(1, 2, 4, 8, 18, 36,72)) +
scale_x_continuous(trans = 'log2', breaks = c(1, 2, 4, 8, 18, 36,72)) +
xlab("Number of Workers") + ylab("Number of Threads")
ggsave(
"timings_MCE_with_OMP.png",
width = 6,
height = 6.5,
units = "in",
dpi = 1200
)
ggplot(datOpenMP, aes(x = Workers, y = Num_threads, z = fitSpaceTime)) +
geom_contour(aes(colour = after_stat(level))) +theme_bw() + theme(
panel.grid.minor = element_blank(),
panel.grid.major = element_blank(),
axis.text.x = element_text(size = 8),
legend.position="bottom"
) +
scale_color_viridis(discrete=FALSE,name="Resource Use ( cp x s)") +
geom_point(data=minSpaceTime,aes(x=Workers,y=Num_threads),color="black",shape="+")+
scale_y_continuous(trans = 'log2', breaks = c(1, 2, 4, 8, 18, 36,72)) +
scale_x_continuous(trans = 'log2', breaks = c(1, 2, 4, 8, 18, 36,72)) +
xlab("Number of Workers") + ylab("Number of Threads")
ggsave(
"spacetime_MCE_with_OMP.png",
width = 6,
height = 6.5,
units = "in",
dpi = 1200
)
################################################################################
## make a copy to replot the statistical model for the data
datOpenMP_log2<-datOpenMP
datOpenMP_log2$Time<-log2(datOpenMP_log2$Time)
datOpenMP_log2$Workers<-log2(datOpenMP_log2$Workers)
datOpenMP_log2$Num_threads<-log2(datOpenMP_log2$Num_threads)
f4_log2<-gam(Time~ti(Workers,bs="ts")+ti(Num_threads)+ti(Workers,Num_threads,bs="ts"),data=datOpenMP_log2)
predTerms<-predict(f4_log2,type="terms",se.fit=TRUE)
datOpenMP_log2$Weff<-predTerms$fit[,1]
datOpenMP_log2$Weffmin<-predTerms$fit[,1]-1.96*predTerms$se.fit[,1]
datOpenMP_log2$Weffmax<-predTerms$fit[,1]+1.96*predTerms$se.fit[,1]
datOpenMP_log2$Teff<-predTerms$fit[,2]
datOpenMP_log2$Teffmin<-predTerms$fit[,2]-1.96*predTerms$se.fit[,2]
datOpenMP_log2$Teffmax<-predTerms$fit[,2]+1.96*predTerms$se.fit[,2]
datOpenMP_log2$WeffTeff<-predTerms$fit[,3]
datOpenMP_log2$WeffTeffmin<-predTerms$fit[,3]-1.96*predTerms$se.fit[,3]
datOpenMP_log2$WeffTeffmax<-predTerms$fit[,3]+1.96*predTerms$se.fit[,3]
png("Worker_Thread_interaction.png",width=6,height=6,units="in",res=1200,pointsize=10)
par(xaxt="n",yaxt="n")
plot(f4_log2,select=3,se=F,rug=FALSE,main = "Worker x Thread Interaction",
xlab = "Workers (log2)",ylab="Threads (log2)")
par(xaxt="s")
axis(1,at=log2(c(1, 2, 4, 8, 18, 36,72)),labels=c(1, 2, 4, 8, 18, 36,72))
par(yaxt="s")
axis(2,at=log2(c(1, 2, 4, 8, 18, 36,72)),labels=c(1, 2, 4, 8, 18, 36,72))
dev.off()
tick<-c(1, 2, 4, 8, 18, 36,72)
ytick<-seq(0,1.5,length.out=5)
pW<-draw(f4,rug=F,se=T,select=1,xlab="Workers (log2)")+
scale_y_continuous(breaks=ytick)+
scale_x_continuous(breaks=log2(tick),labels=tick)+
theme_bw() + theme(
panel.grid.minor = element_blank(),
panel.grid.major = element_blank(),
axis.text.x = element_text(size = 8),
plot.title = element_text(hjust=0.5)
) +
ggtitle("Workers Effect")+
xlab("Workers (log2)") + ylab("Effect on Time (log2 sec)")
( run in 1.494 second using v1.01-cache-2.11-cpan-22024b96cdf )