Adam Shelton
mpi_data = read_csv(here("output_data", "mpi_trials.csv"))
# create line plot with 1/x smoothing
mpi_data %>% subset(hosts == 1) %>% ggplot(aes(x = nodes, y = proc_time/60)) +
geom_smooth(color = color_pal(1, "cool"), size = 1.75, method = "lm",
formula = y ~ I(1/x)) + scale_x_continuous(breaks = c(1,
2, 4, 8, 16)) + labs(title = "Parallelization has Diminishing Returns",
x = "MPI Nodes", y = "Average Running Time (min)") + theme_master(base_size = 22) +
theme(panel.grid.minor.x = element_blank())
# calculate price for each MPI test, converting time to hours
# and requiring two CPUs per node
mpi_data$total_price = 2 * mpi_data$nodes * (mpi_data$proc_time/3600) *
0.031611
# create line plot
mpi_data %>% subset(hosts == 1) %>% ggplot(aes(x = nodes, y = total_price)) +
geom_smooth(color = color_pal(1, "discrete"), size = 1.75,
method = "loess", span = 0.5) + scale_x_continuous(breaks = c(1,
2, 4, 8, 16)) + scale_y_continuous(limits = c(0, 0.02)) +
labs(title = "Parallelization is More Costly", x = "MPI Nodes",
y = "Mean Total Processing Cost (USD)") + theme_master(base_size = 22) +
theme(panel.grid.minor.x = element_blank())
# get means and standard deviations for each number of hosts
mpi_means = mpi_data %>% filter(nodes == 16) %>% filter(hosts ==
1) %>% .$proc_time %>% tibble(hosts = c(1), proc_time = mean(.),
pc_sd = sd(.)) %>% .[1, 2:4]
mpi_means[2, ] = mpi_data %>% filter(nodes == 16) %>% filter(hosts ==
2) %>% .$proc_time %>% tibble(proc_time = mean(.), pc_sd = sd(.)) %>%
.[1, 2:3] %>% bind_cols(host = c(2), .)
mpi_means[3, ] = mpi_data %>% filter(nodes == 16) %>% filter(hosts ==
4) %>% .$proc_time %>% tibble(proc_time = mean(.), pc_sd = sd(.)) %>%
.[1, 2:3] %>% bind_cols(host = c(4), .)
# plot a column plot with error bars
mpi_means %>% ggplot(aes(x = factor(hosts), y = proc_time)) +
geom_col(aes(fill = factor(hosts))) + geom_errorbar(aes(ymin = proc_time -
pc_sd, ymax = proc_time + pc_sd), width = 0.2, position = position_dodge(0.9)) +
scale_fill_manual(values = color_pal(3)) + labs(title = "Number of MPI Hosts Affects Performance",
x = "MPI Hosts", y = "Average Running Time (s)") + theme_master(base_size = 22) +
hide_x_gridlines + hide_legend
# arrange tags by the number of posts for each one
tt_data = read_csv(here("output_data", "top_tags.csv"))
tt_data$count = as.numeric(tt_data$count)
tt_data = tt_data %>% arrange(-count)
tt_data$language = tt_data$language %>% factor()
# take the top six tags and column plot them
tt_data %>% .[1:6, ] %>% ggplot(aes(x = reorder(language, -count),
y = (count/10^6))) + geom_col(aes(fill = language)) + scale_fill_manual(values = color_pal(6)) +
labs(title = "Top Six Tags in StackOverflow Posts", x = "Tag",
y = "Number of Posts (in millions)") + theme_master(base_size = 22) +
hide_x_gridlines + hide_legend
# remove quotation marks and whitespace from each question
ques_year_data = read_csv(here("output_data", "top_questions.csv"))
ques_year_data$question = ques_year_data$question %>% str_remove_all("\"") %>%
str_remove_all("\\\\") %>% unquote(deep = TRUE) %>% str_trim()
ques_year_data = ques_year_data %>% filter(year != 2012) # remove weird results for 2012
# plot flipped axis column plot with questions as labels on
# the plot
ques_year_data %>% ggplot(aes(x = factor(year), y = count)) +
geom_col(aes(fill = year)) + scale_fill_gradientn(colors = color_pal(2)) +
geom_label(aes(x = 1:length(year), y = 110, label = question),
alpha = 0.9, color = "black", family = "Pragati Narrow",
size = 6, hjust = 0) + labs(title = "Top Questions on StackOverflow by Year",
x = "Year", y = "Number of Answers") + theme_master(base_size = 22) +
hide_y_gridlines + hide_legend + coord_flip()
two_grams_data = read_csv(here("output_data", "twograms.csv")) %>%
arrange(-count)
# make graph from top 150 tags
bigram_graph <- two_grams_data[1:150, ] %>% graph_from_data_frame()
remove_axes <- theme(axis.text = element_blank(), axis.line = element_blank(),
axis.ticks = element_blank(), panel.border = element_blank(),
panel.grid = element_blank(), axis.title = element_blank())
# plot a network graph of these top 150 tags
ggraph(bigram_graph, layout = "fr") + geom_edge_link(color = "grey") +
geom_node_point(color = "grey") + geom_node_text(aes(label = name,
color = name %in% tt_data$language[1:6], size = name %in%
tt_data$language[1:6]), nudge_x = 0, nudge_y = 0, repel = TRUE,
family = "Pragati Narrow") + scale_color_manual(values = c("#000000",
color_pal(1, "discrete"))) + scale_size_manual(values = c(5,
8)) + labs(title = "Connections Between Tags in StackOverflow Posts") +
theme_master(base_size = 22) + remove_axes + hide_legend
user_act_data = read_csv(here("output_data", "user_ac_out.csv"))
user_act_data = user_act_data[is.numeric(user_act_data$user_id),
]
# density plot of user activity
ggplot(user_act_data, aes(x = count)) + geom_density(color = color_pal(1,
"cool"), size = 1.75) + labs(title = "Most Users Have Very Little Account Activity",
x = "Account Interactions", y = "Density") + theme_master(base_size = 22)
aggr_ua_data = user_act_data$count %>% table() %>% as.tibble()
names(aggr_ua_data) = c("act_count", "num_obs")
aggr_ua_data$act_count = as.numeric(aggr_ua_data$act_count)
aggr_ua_data$perc = aggr_ua_data$num_obs/sum(aggr_ua_data$num_obs)
# plot treemap of user activity data
ggplot(aggr_ua_data, aes(area = as.numeric(num_obs), fill = as.numeric(act_count),
label = act_count)) + geom_treemap() + geom_treemap_text(colour = "white",
place = "centre", grow = TRUE) + scale_fill_gradientn(trans = "log10",
colors = color_pal(5, type = "continuous")) + labs(title = "The Majority of StackOverflow Accounts Have Very Little Account Activity",
fill = "Number of \nPosts") + theme_master()
user_loc_data = read_csv(here("output_data", "users_gold_badge_locations.csv"))
world_map = map_data("world")
# plot world map with 2d density plot (heatmap) of user
# activity
ggplot() + geom_map(data = world_map, map = world_map, aes(x = long,
y = lat, group = group, map_id = region), fill = "white",
colour = "#7f7f7f", size = 0.5) + stat_density2d(data = user_loc_data,
aes(x = lon, y = lat, fill = stat(level)), geom = "polygon",
alpha = 0.75) + scale_fill_gradientn(colors = color_pal(5,
"continuous", reverse = TRUE)) + theme_map() + theme(legend.position = "right") +
labs(title = "Global Distribution of StackOverflow Users",
fill = "Density") + coord_map("albers", 0, 0)
```