#!/bin/bash

DOTOPTS="-Nfontsize=15 -Tpdf"

dot recycling-noopt.dot $DOTOPTS > recycling-noopt.pdf 

dot recycling-opt.dot $DOTOPTS > recycling-opt.pdf 

dot identity-opt.dot $DOTOPTS > identity-opt.pdf 

dot pushdown-noopt.dot $DOTOPTS > pushdown-noopt.pdf 

dot pushdown-opt.dot $DOTOPTS > pushdown-opt.pdf 

dot parallel.dot $DOTOPTS > parallel-tree.pdf 

dot survey-noopt.dot $DOTOPTS > survey-noopt.pdf 

dot survey-opt.dot $DOTOPTS > survey-opt.pdf 

dot defer.dot $DOTOPTS > defer.pdf 

library(grid)

library(dplyr)

all <- read.csv("results.tsv",sep="\t",header=F, stringsAsFactors=F)

names(all) <- c("exp", "sys", "conf", "s", "r", "timesec")

ybreaks <- c(.01,.1,1,10,60,600)

ylabels <- c("10ms","100ms","1s", "10s","1min", "10min")

xbreaks <- 10^(5:9)

xlabels <- c(expression(10^5),expression(10^6),expression(10^7),expression(10^8),expression(10^9))

theme <- theme_few(base_size = 24) + 

theme(axis.title.y=element_text(vjust=0.9), 

	  axis.title.x=element_text(vjust=-0.1),

	  text=element_text(family="serif"),

	legend.position = "none"

    )

se <- function(x) sqrt(var(x)/length(x))

# parallel

d <- all %>% filter(exp=="parallel", s  == 8) %>% group_by(exp, sys, conf) %>% 

	summarize(meant=mean(timesec), se=se(timesec)) %>% 

	mutate(threads=ifelse(conf=="none", 1L, as.integer(conf))) %>% mutate(tool=paste(sys,threads)) %>% filter(threads < 32)

print(d)

limits <- aes(ymax = meant + se, ymin=meant - se, width=2)

pdf("parallel.pdf",width=10,height=7)

p <- ggplot(d,aes(threads,meant,group=sys)) + 

  geom_point(size=4) + geom_line(size=1.5, aes(group=sys, linetype=sys)) +

  geom_pointrange(limits) +

  scale_y_continuous(limits=c(0,NA)) +

   scale_x_continuous(breaks=unique(d$threads)) +

 geom_vline(xintercept = 10) +

	xlab("Number of Threads") + ylab("Execution Time (s)") + theme +

		annotate("text", x=13.1, y=1.5, label="Problem parallelism", family="serif", size=10) +

print(p)

dev.off()

# f <- d %>% group_by(exp, sys, s) %>% summarize(maxmt=max(meant))

# d <- d %>% left_join(f) %>% mutate(speedup = maxmt/meant) %>% select(s,threads,speedup)

# print(d)

# pdf("speedup.pdf",width=10,height=7)

# p <- ggplot(d,aes(threads,speedup,group=s)) + 

#   geom_point(size=4) + geom_line(size=1.5, aes(group=s, linetype=sys)) +

#   scale_y_continuous(limits=c(0,NA)) +

#    scale_x_continuous(breaks=unique(d$threads)) +

# 	xlab("Number of Threads") + ylab("Speedup") + theme #+

# #	annotate("text", x=10, y=6, label="Problem parallelism", family="serif", size=10) #+

# 	# annotate("text", x=3.5, y=20, label="Renjin ", family="serif", size=10)

# print(p)

# dev.off()

# recycling

d <- all %>% filter(exp=="recycling", s > 5) %>% group_by(exp, sys, s, conf) %>% 

	summarize(meant=mean(timesec), se=se(timesec)) %>% 

	mutate(datasize=10^as.integer(s), tool=paste(sys, conf))

print(d)

limits <- aes(ymax = meant + se, ymin=meant - se, width=2)

pdf("recycling.pdf",width=10,height=7)

p <- ggplot(d,aes(datasize,meant,group=tool)) + 

  geom_point(size=4) + geom_line(size=1.5, aes(group=tool, linetype=tool)) +

  geom_pointrange(limits) +

#  scale_y_log10(breaks=ybreaks, labels=ylabels) +

    scale_x_log10(breaks=xbreaks, labels=xlabels) +

	xlab("Dataset Size (elements, log scale)") + ylab("Execution Time (s)") + theme +

	annotate("text", x=10^7.7, y=65, label="Renjin", family="serif", size=10)+

	annotate("text", x=10^7.7, y=2, label="Renjin + R.", family="serif", size=10)+

	annotate("text", x=10^7.8, y=21, label="GNU R", family="serif", size=10)

print(p)

dev.off()

# # identity

d <- all %>% filter(exp=="identity", s > 5) %>% group_by(exp, sys, conf, s) %>% 

	summarize(meant=mean(timesec), se=se(timesec)) %>% 

	mutate(tool=ifelse(conf=="none", sys, paste(sys,conf)), datasize=10^as.integer(s))

print(d)

limits <- aes(ymax = meant + se, ymin=meant - se, width=2)

pdf("identity.pdf",width=10,height=7)

ggplot(d,aes(datasize,meant,group=tool)) + 

  geom_point(size=4) + geom_line(size=1.5, aes(group=tool, linetype=tool)) +

  geom_pointrange(limits) +

  #scale_y_log10(breaks=ybreaks, labels=ylabels) +

    scale_x_log10(breaks=xbreaks, labels=xlabels) +

	xlab("Dataset Size (elements, log scale)") + ylab("Execution Time (s)") + theme +

	annotate("text", x=60000000, y=4, label="GNU R", family="serif", size=10)+

	annotate("text", x=40000000, y=8.1, label="Renjin ", family="serif", size=10)+

	annotate("text", x=40000000, y=.6, label="Renjin + Identity", family="serif", size=10)

	#scale_color_brewer(palette=cBrwPl) +

 # guides(colour=guide_legend(keywidth=3.5))

dev.off()

# pushdown

	summarize(meant=mean(timesec), se=se(timesec)) %>% 

	mutate(tool=sys, datasize=10^as.integer(s))

print(d)

limits <- aes(ymax = meant + se, ymin=meant - se, width=2)

pdf("pushdown.pdf",width=10,height=7)

p <- ggplot(d,aes(datasize,meant,group=tool)) + 

  geom_point(size=4) + geom_line(size=1.5, aes(group=tool, linetype=tool)) +

  geom_pointrange(limits) +

 # scale_y_log10(breaks=ybreaks, labels=ylabels) +

    scale_x_log10(breaks=xbreaks, labels=xlabels) +

	xlab("Dataset Size (elements, log scale)") + ylab("Execution Time (s)") + theme +

	annotate("text", x=10^7.6, y=.4, label="Renjin ", family="serif", size=10)

print(p)

dev.off()

# operators

d <- all %>% filter(exp=="operators", s > 5) %>% group_by(exp, sys, s, conf) %>% 

	summarize(meant=mean(timesec), se=se(timesec)) %>% 

	mutate(conf=ifelse(conf=="opt"," + Vectorization",""), tool=paste(sys,conf,sep=""), datasize=10^as.integer(s))

print(d)

limits <- aes(ymax = meant + se, ymin=meant - se, width=2)

pdf("operators.pdf",width=10,height=7)

p <- ggplot(d,aes(datasize,meant,group=tool)) + 

  geom_point(size=4) + geom_line(size=1.5, aes(group=tool, linetype=tool)) +

  geom_pointrange(limits) +

   # scale_y_log10(breaks=ybreaks, labels=ylabels) +

    scale_x_log10(breaks=xbreaks, labels=xlabels, limits=c(NA, 10^8.1)) +

source("harness.R")

dd <- function() {as.integer(runif(10^s, 1, 100))}

for (s in 4:sm) {

	a <- list(dd(), dd(), dd(), dd(), dd(), dd(), dd(), dd(), dd(), dd())

	x <- dd()

	dim(x) <- c(length(x),1)

	attr(a, "row.names") <- 1:length(a[[1]])

  	class(a) <- "data.frame"

	for (r in 1:5) {

		timing <- system.time({

			m <- matrix(data=as.numeric(NA), ncol=1, nrow=ncol(a))

			for(i in 1:ncol(a)){

				m[i,]<-t(colSums(a[,i]*x)/sum(a[,i]))

			}

			print(m)

		})[[3]]

		log.result("operators", sys, conf, s, r, timing)

		clearResultRecycler()

	}

}

library(MonetDB.R)

library(survey)

library(sqlsurvey)

options(na.action="na.pass")

#options(monetdb.debug.query=T)

source("harness.R")

for (s in c("alabama", "california", "acs3yr")) { #, "acs3yr"

  svydsgn <- sqlrepsurvey("pwgtp",paste("pwgtp",1:80,sep=""),

    scale=4/80,rscales=rep(1,80), mse=TRUE,

    database="monetdb://localhost/acs2", driver=MonetDB.R(),

    key="idkey",user="monetdb",password="monetdb",table.name=s, check.factors=data.frame())

  for (r in 1:5) {

    timing <- system.time({

      print(svymean(~agep, svydsgn, se=TRUE))

    })[[3]]

    log.result("survey", "sqlsurvey", conf, s, r, timing)

  }

}

# TODO: test with Oct14 branch, probably faster

# TODO: Test on bricks?