big_data_analsis.R

# Plot the data

# Imports
library(reshape2)
library(ggplot2)

# Load the data
# Change path to file
# data_path <- '../parallel/analysis.csv'  # You can use this instead
data_path <- '/home/reynaldo/Documents/School/Spring2017/HPC/project/parallel_mcmc/parallel/complete_big_analysis.csv'
perf_data <- read.csv(file=data_path, header=TRUE, sep=",")
melted_data <- melt(perf_data, id=c("trial", "block", "thread"))
five_hundred_trials <- melted_data[which(melted_data$trial == 500), ]


# Find which combination of blocks and threads works best
five_hundred_trials$thread = as.factor(five_hundred_trials$thread)

ours_five_hundred_trials <- five_hundred_trials[which(five_hundred_trials$variable == "our_time"), ]
ours_five_hundred_trials <- ours_five_hundred_trials[which(!is.na(ours_five_hundred_trials$thread)), ]

line_base <- ggplot(ours_five_hundred_trials, aes(x=block, y=value, group=thread, shape=thread, color=thread)) +
  geom_line() +
  geom_point()
line_base <- line_base +  xlab("Blocks") + ylab("Time (seconds)") + labs(colour = "Num Threads", shape="Num Threads")
line_base <- line_base + ggtitle("Blocks vs Time vs Threads") + theme(plot.title = element_text(hjust = 0.5))
line_base

# Best Number of threads vs The other implementations for Blocks vs Time
five_hundred_trials$thread = as.factor(five_hundred_trials$thread)
one_threads_500_trials <- five_hundred_trials[which(five_hundred_trials$thread == 1), ]

line_base <- ggplot(one_threads_500_trials, aes(x=block, y=value, group=variable, shape=variable, color=variable)) +
  geom_line() +
  geom_point()
line_base <- line_base +  xlab("Blocks") + ylab("Time (seconds)") + labs(colour = "Implementations", shape="Implementations")
line_base <- line_base + ggtitle("Blocks vs Time vs Implementation") + theme(plot.title = element_text(hjust = 0.5))
line_base


# Find which combination of blocks and threads works best
five_hundred_trials$thread = as.factor(five_hundred_trials$thread)
ours_five_hundred_trials <- five_hundred_trials[which(five_hundred_trials$variable == "our_time"), ]
ours_five_hundred_trials <- ours_five_hundred_trials[which(!is.na(ours_five_hundred_trials$thread)), ]

line_base <- ggplot(ours_five_hundred_trials, aes(x=block, y=value, group=thread, shape=thread, color=thread)) +
  geom_line() +
  geom_point()
line_base <- line_base +  xlab("Blocks") + ylab("Time (seconds)") + labs(colour = "Num Threads", shape="Num Threads")
line_base <- line_base + ggtitle("Blocks vs Time vs Threads") + theme(plot.title = element_text(hjust = 0.5))
line_base


# Create Scatter plot (Blocks vs Threads)
five_hundred_trials$block = as.factor(five_hundred_trials$block)

base_plot <- ggplot(five_hundred_trials, aes(x=block, y=time, color=variable))
base_plot <- base_plot  +  geom_point(aes(x=block, y=thread, size=value, color=variable)) 
base_plot <- base_plot + scale_size(range = c(0, 10))
base_plot <- base_plot + xlab("Blocks") + ylab("Time (seconds)") + labs(colour = "Implementation", size="Time")
base_plot <- base_plot + ggtitle("Blocks vs Time vs Implementation") + theme(plot.title = element_text(hjust = 0.5))
base_plot


# Find which combination of blocks and threads works best
ours_five_hundred_trials <- five_hundred_trials[which(five_hundred_trials$variable == "our_time" & five_hundred_trials$block ==1 & five_hundred_trials$thread < 64 ), ]
ours_five_hundred_trials$thread <- as.factor(ours_five_hundred_trials$thread)
ours_five_hundred_trials <- ours_five_hundred_trials[which(!is.na(ours_five_hundred_trials$thread)), ]

bar_base <- ggplot(ours_five_hundred_trials, aes(x=thread, y=value, fill=thread)) + geom_col(width=0.5)
bar_base <- bar_base +  xlab("Threads") + ylab("Time (seconds)")
bar_base <- bar_base + ggtitle("Big Data Threads vs Time (1 Block)") + theme(plot.title = element_text(hjust = 0.5))
bar_base <- bar_base +   theme(legend.position="none")
bar_base