knitr::opts_chunk$set(warning = F, message = F)

This document is an exploration of the killing-blows data recorded from the Iron Gods campaign.

dat <- readr::read_csv(file.path('data', 'ig_events.csv'))
sessions <- readr::read_csv(file.path('data', 'ig_sessions.csv'))

dat <- mutate(
  dat,
  location = factor(location, levels = rev(unique(location))),
  sub_location = factor(
    sub_location, levels = c('Other', rev(unique(sub_location)))),
  chapter = as.factor(chapter),
  monster_type = factor(monster_type, levels = c('Grunt', 'Tough', 'Boss'))
)

sessions$chapter <- as.factor(sessions$chapter)

pal <- c(
  Corrin = "#F6D645FF",
  Hardad = "#4D4D4DFF",
  Nadia = "#DD513AFF",
  Telum = "#35608DFF"
)

# image(as.matrix(1:length(pal)), col = pal)

Session info

We’ve been playing for almost three years! We were pretty active in 2016, less-so in 2017. But 2018 is off to a good start.

sessions %>%
  group_by(as.factor(lubridate::year(session_date))) %>%
  summarize(n = length(unique(session_date))) %>%
  mutate(per_month = round(
    n / c(12, 12, 12, lubridate::month(Sys.Date())), 1)) %>%
  knitr::kable(col.names = c('Year', 'Number of Sessions', 'Sessions per Month'),
               format = 'html') %>%
  kableExtra::kable_styling(full_width = F)

Year	Number of Sessions	Sessions per Month
2015	9	0.8
2016	17	1.4
2017	11	0.9
2018	3	1.5

# Calculate gap info
sessions$gap <- as.numeric(
  sessions$session_date - 
    c(NA, sessions$session_date[1:(nrow(sessions) - 1)]))

longest <- which(sessions$gap == max(sessions$gap, na.rm = T))

The longest gap between sessions was 99 days, between Aug 09 and Nov 16, 2016.

Killing blows

Trends by session

# Create session-level df
time_dat <- dat %>%
  group_by(session_date, chapter) %>%
  summarize(kills = sum(monster_count)) %>%
  full_join(sessions) %>%
  ungroup() %>%
  mutate(chapter = factor(chapter, labels = c(
    'The Fires\nof Creation',
    'Lords of Rust', 
    'The Choking Tower',
    'Valley of the\nBrain Collectors',
    'Palace of\nFallen Stars'))) %>%
  arrange(session_date) %>%
  tidyr::replace_na(list(kills = 0))

Our earliest and more recent sessions were definitely more violent than those in Chapters 3 and 4. Chapter 2, “Lords of Rust,” currently holds the highest body count. However, Chapter 5, “Palace of Fallen Stars,” is on pace to surpass it.

title <- 'Creatures killed by session'

# create labels
chapter_labels <- time_dat %>%
  ungroup() %>%
  filter(!is.na(chapter)) %>%
  mutate(session = as.numeric(as.factor(session_date))) %>%
  group_by(chapter) %>%
  summarize(x = mean(session),
            # x_right = max(session) + 0.5,
            y = max(kills)) %>%
  mutate(
    chapter = factor(chapter, labels = c(
      'The Fires\nof Creation',
      'Lords of Rust', 
      'The Choking Tower',
      'Valley of the\nBrain Collectors',
      'Palace of\nFallen Stars')))

# Offset due to mixed session
# chapter_labels[1, 'x_right'] <- 5
chapter_labels[1, 'x'] <- 2.75
chapter_labels[2, 'x'] <- 7.75
# chapter_labels[nrow(chapter_labels), 'x_right'] <- NA

ggplot(time_dat, aes(x = as.factor(session_date), y = kills, fill = chapter)) +
  geom_bar(stat = 'identity', width = 0.8) +
  geom_text(data = chapter_labels, aes(x, y, label = chapter),
            vjust = -1) +
  # geom_vline(xintercept = chapter_labels$x_right, col = 'gray40') +
  scale_x_discrete(labels = function(x) format.Date(x, '%b %d \'%y')) +
  scale_y_continuous(expand = c(0, 0), limits = c(0, max(time_dat$kills) + 5)) +
  viridis::scale_fill_viridis(discrete = T, option = 'D') +
  guides(fill = F) +
  theme_axes() +
  theme(axis.text.x = element_text(angle = 90, hjust = 1)) +
  labs(x = 'Session', y = 'Killing Blows', title = title)

Trends by player character (PC)

We’ve already seen the overall player data—Corrin and Hardad are leading; Telum and Nadia are behind.

title <- 'Creatures killed by player'

pc_dat <- dat %>%
  filter(!is.na(pc)) %>%
  group_by(pc) %>%
  summarize(kills = sum(monster_count, na.rm = T))

ggplot(pc_dat, aes(x = pc, y = kills, label = kills, fill = pc)) +
  geom_bar(stat = 'identity') +
  geom_label(fill = 'white') +
  scale_y_continuous(expand = c(0, 0), limits = c(0, max(pc_dat$kills) + 5)) +
  scale_fill_manual("PC", values = pal) +
  guides(fill = F) +
  labs(x = 'Player Character', y = 'Killing Blows') +
  theme_axes() + theme(axis.text.x = element_text(size = 12))

However, that’s been changing over time. The biggest change has been Hardad, who started off trailing the party, caught up to Telum and Nadia by 6th level, and has been outpacing Corrin for the past few levels.

Use the tabs to explore!

# Group data for next few plots (pc / session)
pc_dat <- dat %>%
  group_by(session_date, pc) %>%
  summarize(kills = sum(monster_count)) %>%
  full_join(sessions) %>%
  arrange(session_date) %>%
  tidyr::replace_na(list(kills = 0)) %>%
  ungroup() %>%
  mutate(pc = factor(pc, levels = unique(pc)))

Total kills per level

title <- 'Total kills by PC per level'

dat %>%
  group_by(pc_level, pc) %>%
  summarize(kills = sum(monster_count)) %>%
  filter(!is.na(pc) & pc_level < 14) %>%
  group_by(pc) %>%
  arrange(pc_level) %>%
  mutate(total_kills = cumsum(kills)) %>%
  ggplot(aes(x = as.factor(pc_level), y = total_kills, color = pc, group = pc)) +
  geom_line() + geom_point() +
  scale_y_continuous(expand = c(0, 0)) +
  scale_color_manual("PC", values = pal) +
  guides(color = F) +
  annotate('text', x = c(6, 9, 11, 12), y = c(75, 65, 55, 30), 
           label = c('Corrin', 'Hardad', 'Telum', 'Nadia')) +
  theme_axes() +
  labs(x = 'PC Level', y = 'Total Kills', title = title)

Total kills over time

title <- 'Total kills by PC per session'
pc_dat %>%
  filter(!is.na(pc)) %>%
  group_by(pc) %>%
  arrange(session_date) %>%
  mutate(total_kills = cumsum(kills)) %>%
  ggplot(aes(x = as.factor(session_date), y = total_kills, color = pc, group = pc)) +
  geom_line() + geom_point() +
  scale_color_manual("PC", values = pal) +
  guides(color = F) +
  annotate('text', x = c(15, 20, 25, 30), y = c(85, 60, 50, 30), 
           label = c('Corrin', 'Hardad', 'Telum', 'Nadia')) +
  scale_x_discrete(labels = function(x) format.Date(x, '%b %d \'%y')) +
  scale_y_continuous(expand = c(0, 0)) +
  theme_axes() +
  theme(axis.text.x = element_text(angle = 90, hjust = 1)) +
  labs(x = 'Session', y = 'Total kills', title = title)

Kills per session

title <- 'Kills by PC per session'
max_kills <- pc_dat %>% group_by(session_date) %>% 
  summarize(kills = sum(kills)) %>% 
  select(kills) %>% drop %>% max(na.rm = T)

pc_dat %>% filter(!is.na(pc)) %>%
  ggplot(aes(x = as.factor(session_date), y = kills, fill = pc)) +
  geom_bar(stat = 'identity', width = 0.8) +
  scale_x_discrete(labels = function(x) format.Date(x, '%b %d \'%y')) +
  scale_y_continuous(expand = c(0, 0), limits = c(0, max_kills + 1)) +
  scale_fill_manual("PC", values = pal) +
  theme_axes() +
  theme(axis.text.x = element_text(angle = 90, hjust = 1)) +
  labs(x = 'Session', y = 'Kills', title = title)

Trends by location

Use the tabs to explore!

Geographic Location

The party definitely left more bodies behind at particular locations. In general, less-civilized but populous locations see higher death counts. Scrapwall was particularly brutal.

title <- 'Kills by location'

dat %>%
  filter(!is.na(pc)) %>%
  group_by(location, pc) %>%
  summarize(kills = sum(monster_count)) %>%
  ggplot(aes(x = location, y = kills, fill = pc)) +
  geom_bar(stat = 'identity') +
  scale_fill_manual("PC:", values = pal, guide = guide_legend(reverse = T)) + 
  scale_y_continuous(expand = c(0, 0), limits = c(0, 105)) +
  coord_flip() +
  labs(y = 'Creatures Killed', x = '', title = title) +
  theme_axes() +
  theme(axis.text.y = element_text(size = 12),
        legend.position = 'top')

Specific sites

However, the single most deadly place we’ve been is definitely the Black Sovereign’s Palace, which just goes to show that even with crazy dangerous wilderness regions and lawless settlements, political struggles are the most violent.

title <- 'Kills by specific site'
dat %>%
  filter(!is.na(pc)) %>%
  mutate(location = factor(location, levels = unique(location))) %>%
  # tidyr::replace_na(list(sub_location = 'Other')) %>%
  group_by(location, sub_location, pc) %>%
  summarize(kills = sum(monster_count)) %>%
  ggplot(aes(x = sub_location, y = kills, fill = pc)) +
  geom_bar(stat = 'identity') +
  scale_y_continuous(expand = c(0, 0), limits = c(0, 58)) +
  coord_flip() +
  facet_grid(location ~ ., scales = 'free_y', space = 'free', switch = 'y') +
  theme_axes() +
  theme(strip.text.y = element_text(angle = 180),
        legend.position = 'top') +
  scale_fill_manual("PC:", values = pal, guide = guide_legend(reverse = T)) +
  labs(x = '', y = 'Monsters Killed', title = title)

Who’s better where?

Different PCs stand out in different locations. Excluding locations with fewer than 10 kills, where did each PC stand out the most - that is, where did they claim their highest proportion of the party’s kills?

top_pc <- dat %>%
  group_by(location, sub_location, pc) %>%
  summarize(pc_kills = sum(monster_count)) %>%
  group_by(location, sub_location) %>%
  mutate(location_kills = sum(pc_kills),
         pc_kill_pct = pc_kills / location_kills) %>%
  filter(location_kills >= 10) %>%
  group_by(pc) %>%
  arrange(desc(pc_kill_pct)) %>%
  slice(1:3) %>%
  select(pc, location, sub_location, location_kills, pc_kills, pc_kill_pct) %>%
  mutate(pc_kill_pct = scales::percent(round(pc_kill_pct, 2)))

# ggplot(top_pc, aes(x = sprintf('%s, %s', sub_location, location),
#                    y = pc_kill_pct,
#                    fill = pc)) +
#   geom_bar(stat = 'identity') +
#   facet_wrap(~ pc, scales = 'free_y', ncol = 1) +
#   scale_y_continuous(labels = scales::percent) +
#   scale_fill_manual(values = pal) +
#   guides(fill = FALSE) +
#   labs(x = 'Percent', y = '') +
#   theme_bw() +
#   coord_flip()

view_top <- function(dat, char){
  dat %>%
    filter(pc == char) %>%
    knitr::kable(
      col.names = c('PC', 'Location', 'Site', 'Party kills', 
                    sprintf('%s kills', char), 'Proportion'),
      align = c('lllrrr')) %>%
    print()
}

Corrin stood out most fighting the Dominion, and at the beginning of the game, where starting with a two-handed weapon and high strength gives a big advantage.

top_pc %>% view_top('Corrin')

PC	Location	Site	Party kills	Corrin kills	Proportion
Corrin	Scar of the Spider	Dominion Dropship	17	8	47%
Corrin	Scar of the Spider	Dominion Hive	33	14	42%
Corrin	Torch	Buried Ship	45	19	42%

Hardad stood out most fighting in the Fungal Caves, where we encountered many flying enemies, as well as taking all the kills Corrin didn’t in the Dominion Dropship. More recently, he cleaned up at the Black Sovereign’s Palace, a target-rich environment. Note: Hardad also killed nearly every enemy in the Machine Caves with his EMP pistol, but that was only 8 enemies total and was dropped from this analysis.

top_pc %>% view_top('Hardad')

PC	Location	Site	Party kills	Hardad kills	Proportion
Hardad	Starfall	Technic League Compound	16	12	75%
Hardad	Scar of the Spider	Fungal Caves	21	9	43%
Hardad	Starfall	Black Sovereign’s Palace	55	23	42%

Telum stood out most in the ship under Torch, where he was the only early-game source of electric damage for the party. He also did very well in Xoud’s Tower, where he claimed a higher-than-usual number of kills against the robots and weird magical creatures (including Xoud himself!).

top_pc %>% view_top('Telum')

PC	Location	Site	Party kills	Telum kills	Proportion
Telum	Torch	Buried Ship	45	14	31%
Telum	Smokewood	Choking Tower	26	7	27%
Telum	Scrapwall	Smilers’ Hideout	30	5	17%

Nadia stood out most in the early portions of Scrapwall, where there were numerous foes weak to blasting spells and grenades. She also increased her body count a lot at the Black Sovereign’s Palace, facing many weaker enemies with new metamagic.

top_pc %>% view_top('Nadia')

PC	Location	Site	Party kills	Nadia kills	Proportion
Nadia	Scrapwall	Birdfoot’s Hideout	15	5	33%
Nadia	Starfall	Black Sovereign’s Palace	55	15	27%
Nadia	Scrapwall	Smilers’ Hideout	30	7	23%

Trends by weapon

Each PC has 1-2 clear standout weapons, with Hardad bringing the most diversity to the team. Some surprises include that Nadia has killed more enemies with a hammer than with her summons, Hardad’s killed more people with Holy Water than with the Arc Pistol, and Corrin’s chainsaw only just surpassed the greatsword around level 13.

Overall

title <- 'Killing blows by weapon'
dat %>%
  filter(!is.na(pc)) %>%
  group_by(kill_method, pc) %>%
  summarize(kills = sum(monster_count)) %>%
  arrange(kills) %>%
  ungroup() %>%
  mutate(kill_method = factor(kill_method, levels = unique(kill_method))) %>%
  ggplot(aes(x = kill_method, y = kills, fill = pc)) +
  geom_col() +
  facet_grid(pc ~ ., scales = 'free_y', space = 'free',
             switch = 'y') +
  scale_y_continuous(expand = c(0, 0)) +
  scale_fill_manual(values = pal) + guides(fill = F) +
  theme_axes() +
  coord_flip() +
  labs(y = 'Killing blows', x = '', title = title)

Per session

title <- 'Kills by weapon per session'

max_kills <- dat %>% group_by(kill_method) %>% 
  summarize(kills = sum(monster_count)) %>%
  select(kills) %>% drop() %>% max()

labs <- dat %>% 
  group_by(pc, kill_method) %>% 
  arrange(pc_level) %>%
  summarize(pc_level = max(pc_level),
            kills = sum(monster_count)) %>%
  filter(kills > 5)

killdat <- dat %>% 
  filter(!is.na(pc)) %>%
  group_by(pc_level, pc, kill_method) %>%
  summarize(kills = sum(monster_count)) %>%
  group_by(pc, kill_method) %>%
  arrange(pc_level) %>%
  mutate(total_kills = cumsum(kills)) %>%
  ungroup()

for(i in unique(killdat$kill_method)){
  temp <- subset(killdat, kill_method == i)
  killdat <- rbind(
    killdat, 
    tibble(pc_level = min(temp$pc_level) - 1,
           pc = unique(temp$pc),
           kill_method = i,
           kills = 0,
           total_kills = 0))
}

ggplot(killdat, 
       aes(x = pc_level, y = total_kills, 
           color = pc, group = kill_method)) +
  geom_line() + geom_point() +
  ggrepel::geom_label_repel(
    data = labs, 
    aes(x = pc_level, y = kills, label = kill_method),
    show.legend = F) +
  scale_y_continuous(expand = c(0, 0), limits = c(0, max_kills + 5)) +
  scale_color_manual("PC", values = pal) +
  theme_axes() +
  labs(x = 'Level', y = 'Kills', title = title)

Creature Type

Not all enemies are created equal. We’ve categorized the creatures we faced into three types: Grunts, Toughs, and Bosses.

dat %>% group_by(monster_type) %>% summarize(kills = sum(monster_count)) %>%
  knitr::kable(col.names = c('Creature Type', 'Count'), format = 'html') %>%
  kableExtra::kable_styling(full_width = F)

Creature Type	Count
Grunt	212
Tough	125
Boss	32
NA	4

Creature Type by Geography

This can give us a rough breakdown of the kind of enemies we faced in each location. Were they fewer and tougher, or more numerous and weaker?

Location

Scrapwall had the highest proportion of Grunts, although Torch and Iadenveigh are close seconds. Smokewood stands out for its tougher enemies; the Scar of the Spider and Starfall are somewhere in the middle.

title <- 'Creature Breakdown by Location'
dat %>%
  filter(!is.na(monster_type)) %>%
  group_by(location) %>% filter(n() > 10) %>%
  mutate(monster_type = factor(
    monster_type, 
    levels = rev(levels(monster_type)))) %>%
  group_by(monster_type, location) %>%
  summarize(kills = sum(monster_count)) %>%
  group_by(location) %>%
  mutate(kills_pct = kills / sum(kills)) %>%
  ggplot(aes(x = location, y = kills_pct, fill = monster_type)) +
  geom_bar(stat = 'identity') +
  scale_y_continuous(
    labels = scales::percent, expand = c(0, 0), limits = c(0, 1.05)) +
  scale_fill_brewer('Creature Type:', palette = 7, direction = -1,
                    guide = guide_legend(reverse = T)) +
  labs(y = 'Creatures', x = '', title = title) +
  theme_blank() +
  theme(legend.position = 'top',
        axis.text.y = element_text(size = 12)) +
  coord_flip()

Specific sites

However, there is variation within locations. The Dominion Dropship, for example, has the highest proportion of Tough enemies of any major site in the game, and all of the Starfall sites outside of the Black Sovereign’s palace were tougher enemies.

title <- 'Creature Breakdown by Specific Site'
dat %>%
  filter(!is.na(monster_type)) %>%
  mutate(monster_type = factor(
    monster_type, 
    levels = rev(levels(monster_type)))) %>%
  group_by(location) %>% filter(n() > 10) %>%
  ungroup() %>%
  mutate(location = factor(location, levels = unique(location))) %>%
  # tidyr::replace_na(list(sub_location = 'Other')) %>%
  group_by(monster_type, location, sub_location) %>%
  summarize(kills = sum(monster_count)) %>%
  group_by(location, sub_location) %>%
  mutate(kills_pct = kills / sum(kills)) %>%
  ggplot(aes(x = sub_location, y = kills_pct, fill = monster_type)) +
  geom_bar(stat = 'identity') +
  coord_flip() +
  facet_grid(location ~ ., scales = 'free_y', space = 'free',
             switch = 'y') +
  scale_y_continuous(labels = scales::percent, 
                     expand = c(0, 0), limits = c(0, 1.05)) +
  scale_fill_brewer(
    '', palette = 7, direction = -1, guide = guide_legend(reverse = T)) +
  labs(y = 'Creatures', x = '', title = title) +
  theme_blank() +
  theme(legend.position = 'top',
        strip.text.y = element_text(angle = 180))

Creature Types and PCs

Number of kills by PC

Some PCs are better at dealing with certain kinds of monsters. Corrin has killed the most Bosses and Grunts, but Hardad has killed the most Toughs.

title <- 'Types of creatures killed by PCs'
dat %>%
  filter(!is.na(monster_type)) %>%
  group_by(pc, monster_type) %>%
  summarize(kills = sum(monster_count)) %>%
  ggplot(aes(x = pc, y = kills, fill = monster_type)) +
  geom_bar(stat = 'identity', position = 'dodge') +
  scale_y_continuous(expand = c(0, 0), limits = c(0, 85)) +
  scale_fill_brewer('Creature Type:', palette = 7) +
  theme_axes() + theme(axis.text.x = element_text(size = 12)) +
  labs(x = 'Player Character', y = 'Creatures Killed', title = title)

Proportion of kills by PC

As a proportion of their kills, Telum actually has the most Boss-heavy list, while Nadia is most heavily skewed towards killing Grunts.

title <- 'Proportion of PC\'s kills by type'
dat %>%
  filter(!is.na(monster_type)) %>%
  mutate(monster_type = factor(
    monster_type, 
    levels = rev(levels(monster_type)))) %>%
  group_by(pc, monster_type) %>%
  summarize(kills = sum(monster_count)) %>%
  group_by(pc) %>%
  mutate(kills_pct = kills / sum(kills)) %>%
  ggplot(aes(x = pc, y = kills_pct, fill = monster_type)) +
  geom_bar(stat = 'identity') +
  scale_y_continuous(labels = scales::percent, expand = c(0, 0), limits = c(0, 1.05)) +
  scale_fill_brewer('Creature Type:', palette = 7, direction = -1,
                    guide = guide_legend(reverse = T)) +
  theme_blank() + 
  theme(axis.text.y = element_text(size = 12),
        legend.position = 'top') +
  coord_flip() +
  labs(x = '', y = 'Creatures Killed', title = title)

Changes by PC over time

Telum’s been killing bosses this whole time, but otherwise I think the main takeaway here is that we’re coding more enemies as Toughs as we enter higher levels. That might be a tendency to code enemies that were previously Toughs as if they’re still Toughs, a bit longer than we should.

title <- 'Proportion of PC\'s kills by type over time'
dat %>%
  filter(!is.na(monster_type)) %>%
  mutate(pc_level = factor(pc_level, levels = max(pc_level, na.rm = T):1),
         monster_type = factor(
           monster_type, 
           levels = rev(levels(monster_type)))) %>%
  group_by(pc, pc_level, monster_type) %>%
  summarize(kills = sum(monster_count)) %>%
  group_by(pc, pc_level) %>%
  mutate(kills_pct = kills / sum(kills)) %>%
  ggplot(aes(x = pc_level, y = kills_pct, fill = monster_type)) +
  geom_bar(stat = 'identity') +
  facet_grid(pc ~ ., switch = 'y') +
  scale_x_discrete(expand = c(0, 0)) +
  scale_y_continuous(labels = scales::percent, expand = c(0, 0)) +
  coord_flip() +
  scale_fill_brewer(
    'Creature Type:', palette = 7, direction = -1,
    guide = guide_legend(reverse = T)) +
  theme_bw() + 
  theme(axis.text.y = element_text(size = 12),
        strip.text = element_text(angle = 180),
        panel.spacing = unit(10, "pt"),
        legend.position = 'top') +
  labs(x = '', y = 'Creatures Killed', title = title)

Perception of difficulty

The “objective” measure of a creature’s difficulty is its Challenge Rating (CR), which is what Paizo uses to balance encounters. However, some fights prove more difficult than others, because of PC specializations, terrain, spells used to date, or just crazy dice.

Changes over time

Comparing PC level to CR, we normally assess an enemy as “Tough” if it has a CR roughly equal to the PC’s level, a “Boss” if its CR is higher than our level, and a “Grunt” if its CR is lower than our level. This is consistent over time.

However, as the graph below indicates, we’ve been increasingly willing to code enemies as “Tough” even if their CR is well below our own. That’s not necessarily because we’re finding fights more difficult—rather, we may be encountering enemies we’ve previously coded as Toughs, and coding them as Toughs again, rather than updating their difficulty to Grunt. Enemies with CR above ours—who are often unique, one-time foes—are consistently coded as Bosses.

title <- 'Creature CR versus PC Level'
subtitle <- 'Lines represent the trend in our assesments over time'

lab <- data.frame(
  x = c(11, 12, 5),
  y = c(3, 10, 8),
  text = factor(c('Grunt', 'Tough', 'Boss'), 
                levels = c('Grunt', 'Tough', 'Boss')),
  color = c('black', 'white', 'white'), 
  fill = viridis::viridis(3, end = 0.9, direction = -1),
  stringsAsFactors = F)

dat %>%
  filter(monster_cr != 0 & !is.na(monster_type)) %>%
  ggplot(aes(x = pc_level, y = monster_cr, color = monster_type)) +
  geom_abline(slope = 1, intercept = 0, linetype = 'dashed') +
  geom_smooth(method = 'lm', se = F, alpha = 0.75) +
  geom_count(alpha = 0.75, position = position_dodge(width = 0.5)) +
  geom_label(dat = lab, alpha = 0.35, color = 'black',
             aes(x = x, y = y, label = text, fill = text)) +
  viridis::scale_color_viridis(
    discrete = T, direction = -1, end = 0.9) +
  viridis::scale_fill_viridis(
    discrete = T, direction = -1, end = 0.9) +
  guides(color = F, fill = F) +
  scale_size('Number of Creatures', breaks = c(1, 2, 5, 10)) +
  scale_y_continuous(
    expand = c(0, 0.1), limits = c(0, 15), breaks = 1:15) +
  scale_x_continuous(
    expand = c(0, 0.1), limits = c(0, 13), breaks = 1:13) +
  coord_fixed() + theme_axes() + theme(legend.position = 'top') +
  labs(x = 'PC Level', y = 'Challenge Rating', title = title, subtitle = subtitle)

Easier than expected

These are enemies that we either

Coded as ‘Tough’ even though their CR was higher than ours, or
Coded as ‘Grunt’ even though their CR was close to ours (CR over 3/4 of PC level)

dat %>%
  filter(monster_cr != 0) %>%
  mutate(cr_level = monster_cr / pc_level) %>%
  filter(
    (monster_type == 'Tough' & cr_level > 1) | 
      (monster_type == 'Grunt' & cr_level > 0.75)) %>%
  select(monster_cr, pc_level, monster_name, monster_type) %>%
  unique() %>%
  knitr::kable(col.names = c('Creature CR', 'PC Level', 
                             'Creature Name', 'Assessed Difficulty'))

Creature CR	PC Level	Creature Name	Assessed Difficulty
2	1	Blindheim	Tough
1	1	Rat	Grunt
1	1	Weird mutant rat	Grunt
1	1	Gremlin	Grunt
2	1	Gremlin leader	Tough
2	1	Repair Drone	Grunt
2	1	Brown mold	Grunt
3	2	Cerebral fungus	Grunt
3	2	Vegepygmy boss	Tough
2	2	Doctor robot	Grunt
4	3	Gearsman	Tough
6	5	Will-o’ Wisp	Tough
7	6	Chuul	Tough
8	7	Fancy Looking Gearsman	Tough
9	8	Deathtrap Ooze	Tough
7	9	Nanite Swarm	Grunt
7	9	Belker	Grunt
9	10	Tentacled Plant Mass	Grunt
12	11	Dolphin-Eel thing	Tough
12	11	N. Gin Attendant	Tough

Harder than expected

These are enemies who we either

Coded as ‘Boss’ even though they had CR at or below us, or
Coded as ‘Tough’ even though they had CR well below ours (less than 3/4 of PC level)

dat %>%
  filter(monster_cr != 0) %>%
  mutate(cr_level = monster_cr / pc_level) %>%
  filter(
    (monster_type == 'Boss' & cr_level <= 1) |
      (monster_type == 'Tough' & cr_level < 0.75)) %>%
  select(monster_cr, pc_level, monster_name, monster_type) %>%
  unique() %>%
  knitr::kable(col.names = c('Creature CR', 'PC Level', 
                             'Creature Name', 'Assessed Difficulty'))

Creature CR	PC Level	Creature Name	Assessed Difficulty
3	4	Sanvil Trett	Boss
3	4	Garmin Ulreth	Boss
5	5	Marrow	Boss
3	6	Ogre	Tough
3	6	Nalakai	Tough
6	6	Draig	Boss
6	9	Warhammer Wobot	Tough
6	9	Achaierai	Tough
4	9	Allip	Tough
6	9	Hungry Fog	Tough
7	10	Shambling Mounds	Tough
6	10	Mi-Go	Tough
6	11	Mi-Go	Tough
11	11	Pale Stranger	Boss
8	11	Neh-Thalggu	Tough
8	11	Intellect Devourer	Tough
6	12	Floating Gray Orb?	Tough
12	12	Possessed Ebony-Green Dragon	Boss
8	12	Neh-Thalggu	Tough
12	12	Paajet	Boss

Classification

Just for fun, what if we use “\(k\)-nearest neighbors” classification to model what how the party will classify enemies?

# Make dataset with no missing values
knn_train <- dat %>% 
  filter(!is.na(monster_cr) & !is.na(pc_level) & !is.na(monster_type)) %>%
  select(monster_cr, pc_level, monster_type)

# Run models for all k 1-20 and find the one with the best in-sample classification
# note, this is dumb because it over-fits the data
results <- rep(NA, 20)
for(k in seq_along(results)){
    mod <- class::knn.cv(
      knn_train[, c('monster_cr', 'pc_level')], 
      cl = knn_train$monster_type, 
      k = k)
    results[k] <- mean(knn_train$monster_type == mod)
}

# What k is best?
which(results == max(results))

## [1] 2

# If there are more than one, pick the first / lowest k
k <- which(results == max(results))[1]

# Predict for all combinations of monster CR and pc level
grid <- na.omit(expand.grid(
  unique(round(dat$monster_cr)),
  unique(dat$pc_level)))

mod <- class::knn(
  train = knn_train[, c('monster_cr', 'pc_level')], 
  test = grid, 
  cl = knn_train$monster_type, 
  # set k to the model with the best in-sample prediction
  k = k)

data.frame(class = mod, x = grid[, 2], y = grid[, 1]) %>%
  ggplot(aes(x = x, y = y, color = class)) + 
  geom_abline(slope = 1, intercept = 0, color = 'gray', size = 2) + 
  geom_point(size = 5, shape = 15) + theme_bw() + 
  viridis::scale_color_viridis('Assessed\nDifficulty', discrete = T, direction = -1) +
  coord_fixed() +
  labs(x = 'PC Level', y = 'Creature CR', title = 'Predicted Player Classification',
       subtitle = 'Based on "k-nearest neighbors" classification with k = 2')

Fun stuff

Alignment Chart

Heard you like charts so we made a chart of your alignment chart so you can categorize while you moralize

dat$monster_alignment[dat$monster_alignment == 'N'] <- 'NN'

dat <- dat %>%
  mutate(align_order = substring(monster_alignment, 1, 1),
         align_order = factor(align_order, levels = c('L', 'N', 'C'),
                              labels = c('Lawful', 'Neutral', 'Chaotic')),
         align_moral = substring(monster_alignment, 2, 2),
         align_moral = factor(align_moral, levels = c('E', 'N', 'G'),
                              labels = c('Evil', 'Neutral', 'Good')))


dat_melt <- dat[rep(1:nrow(dat), dat$monster_count), 
                c('pc', 'monster_cr', 'align_order', 'align_moral')]

jit <- 0.35
dat_melt %>% filter(!is.na(align_moral) & !is.na(pc)) %>%
  ggplot(aes(x = align_order, y = align_moral, size = monster_cr, color = pc)) +
  geom_point(alpha = 0.75, position = position_jitter(width = jit, height = jit)) +
  scale_x_discrete(drop = F) +
  scale_y_discrete(drop = F) +
  scale_size('Challenge Rating') +
  scale_color_manual('Player Character', values = pal) +
  geom_hline(yintercept = c(2.5, 1.5)) + 
  geom_vline(xintercept = c(2.5, 1.5)) +
  coord_fixed() +
  theme_bw() +
  theme(panel.grid = element_blank(), 
        axis.text = element_text(size = 12)) +
  labs(x = '', y = '')

Boss list!

Let’s not forget our dear departed Bosses!

dat$monster_name[grepl('BINOX', dat$monster_name)] <- 'Binox the Builder'

dat %>% 
  filter(monster_type == 'Boss') %>% 
  select(monster_name, pc, sub_location, kill_method) %>%
  knitr::kable(col.names = c(
    'This creature...', 'was killed by...', 'in the...', 'with the...'
  ))

This creature…	was killed by…	in the…	with the…
Four Armed Zombie (Boss)	Hardad	Buried Ship	Holy Water
Meyanda the Android	Telum	Buried Ship	Shocking Grasp
Sanvil Trett	Corrin	City of Torch	Greatsword
Garmin Ulreth	Telum	City of Torch	Color Spray
Birdfoot	Corrin	Birdfoot’s Hideout	Greatsword
Marrow	Telum	Smilers’ Hideout	Shocking Grasp
Mutant Mantacore	Corrin	Scrapwall	Longbow
Wraith	Corrin	Haunted Canyon	Greatsword
Helskaarg	Corrin	Arena	Greatsword
Kulgara	Telum	Hellion’s Lair	Scimitar
Draig	Hardad	Hellion’s Lair	Magma Pistol
Helion	Hardad	Hellion’s Lair	EMP Pistol
Varisian Technic League Spy	Corrin	Android Factory	Chainsaw
Giant Excavator Robot	Corrin	Smokewood	Chainsaw
Cavenenchin the Leukodaemon	Corrin	Choking Tower	Chainsaw
Furkus Xoud	Telum	Choking Tower	Shocking Grasp
8 Legged Robot	Hardad	Choking Tower	EMP Pistol
Binox the Builder	Telum	Machine Caves	Shocking Grasp
Rainbow Mi-Go	Hardad	Fungal Caves	Pistol
Pale Stranger	Corrin	Scar of the Spider	Chainsaw
Froghemoth	Hardad	Scar of the Spider	Laser Pistol
The-Stars-Whispers	Corrin	Dominion Dropship	Chainsaw
Annihilator Robot	Corrin	Dominion Hive	Chainsaw
Possessed Ebony-Green Dragon	Hardad	Dominion Hive	Laser Pistol
Paajet	Telum	Dominion Hive	Shocking Grasp
Giant Chitinous Monster	Nadia	Dominion Hive	Flame Strike
Doc Hellbroth	Corrin	Red Reaver	Chainsaw
Technic League Captain	Corrin	Night Market	Chainsaw
Kevoth Kul, The Black Sovereign	Telum	Black Sovereign’s Palace	Scimitar
Tek Makul (Tech McCool)	Telum	Black Sovereign’s Palace	Frigid Touch
Ghartone (We think)	Telum	Black Sovereign’s Palace	Shocking Grasp
Dybbuk	Corrin	Black Sovereign’s Palace	Holy Chainsaw

Notes

Sometimes we add notes to the sheet. That makes the following entries… notable.

dat %>%
  filter(!is.na(notes)) %>%
  select(monster_name, pc, sub_location, kill_method, notes) %>%
  knitr::kable(col.names = c(
    'Creature', 'PC', 'Location', 'Method', 'Notes'
  ))

Creature	PC	Location	Method	Notes
Three-legged robot	Corrin	City of Torch	Greatsword	FIRST BLOOD
Blindheim	Hardad	Caves	Grit	Up Close and Personal after it blinded half the party
Rat	Nadia	Caves	Burning Hands	Two-in-one!
Weird mutant rat	Corrin	Caves	Greatsword	CRITICAL HIT
Green Slime	Telum	Buried Ship	Scimitar	First Kill!
Four-armed undead thing	Corrin	Buried Ship	Greatsword	Foul monstrosities!
Four-armed undead thing	Hardad	Buried Ship	Holy Water	3 throws, 3 kills
Four-armed undead thing	Telum	Buried Ship	Scimitar	Flank!
Flying Fungus	Hardad	Buried Ship	Grit	Up Close and Personal after it paralyzed Nadia
Four Armed Zombie (Boss)	Hardad	Buried Ship	Holy Water	Missed but the 1 damage was enough!
Vegepygmy minion	Corrin	Buried Ship	Greatsword	Attack of opportunity
Vegepygmy minion	Telum	Buried Ship	Scimitar	Critical hit!
Vegepygmy boss	Corrin	Buried Ship	Greatsword	Natural 20
Doctor robot	Telum	Buried Ship	Shocking Grasp	Critical + electric vulnerability
Flying robot	Telum	Buried Ship	Shocking Grasp	Critical + electric vulnerability… AGAIN
Doctor robot	Telum	Buried Ship	Scimitar	After going down and getting anesthetized
Giant tentacle monster	Nadia	Buried Ship	Crossbow bolt	Sniped in melee combat!
Half-orc	Corrin	Buried Ship	Greatsword	AoO from Nadia’s “FLEE” command
Ratfolk	Corrin	Buried Ship	Greatsword	Charge
Half-orc	Telum	Buried Ship	Color Spray	Knocks them out, to be killed
Half-orc	Telum	Buried Ship	Scimitar	The half-orc that wouldn’t die
Thug	Corrin	City of Torch	Greatsword	This was, admittedly, pretty unprovoked
Garmin Ulreth	Telum	City of Torch	Color Spray	Not really dead, just subdued
Hatchet Bandit	Nadia	Aldonard’s Grave	Burning Hands	An act of incredible valor
Birdfoot’s Orc	Corrin	Birdfoot’s Hideout	Greatsword	2 if by crit
Birdfoot’s Orc	Nadia	Birdfoot’s Hideout	Burning Hands	1 by smoke inhalation
Birdfoot	Corrin	Birdfoot’s Hideout	Greatsword	After Birdfoot murderizes Hardad
Kij the Bird	Nadia	Birdfoot’s Hideout	Hammer	Coup de Grace
Pile of Moving Junk	Hardad	Dinvaya’s Hideout	Pistol	Crit
Smiler	Corrin	Smilers’ Hideout	Greatsword	Saving Ratfolk
Smiler	Nadia	Smilers’ Hideout	Burning Hands	After a well-placed grenade
Human Corpse	Nadia	Smilers’ Hideout	Fireball	Nadia’s Very First Fireball
Lobotomites	Corrin	Smilers’ Hideout	Greatsword	Cleaving majestically
Smiler	Corrin	Smilers’ Hideout	Greatsword	Dramatically appropriate crit
Smiler	Corrin	Smilers’ Hideout	Longbow	Did I mention he killed him with a long bow.
Mutant Mantacore	Corrin	Scrapwall	Longbow	WTF with this longbow! (stole from Telum)
Wraith	Corrin	Haunted Canyon	Greatsword	Crit
The Dish	Hardad	Scrapwall	Cylex	And boom goes the dynamite
Helion	Hardad	Hellion’s Lair	EMP Pistol	Crit!
Yauguai	Corrin	Farm	Chainsaw	First chainsaw kill!
Deformed Andoid	Corrin	Android Factory	Chainsaw	SO MANY CRITS
Deformed Andoid	NA	Android Factory	NA	Blew herself up
Varisian Technic League Spy	Corrin	Android Factory	Chainsaw	But she didn’t die, because we saved her (only to kill her later?)
Gearsman	Hardad	Android Factory	EMP Pistol	It’s super effective!
Gearsman	Telum	Android Factory	Shocking Grasp	OVERKILL
Warhammer Wobot	Hardad	Choking Tower	EMP Pistol	After 3 of 4 pass out!
Cavenenchin the Leukodaemon	Corrin	Choking Tower	Chainsaw	Perfectly timed crit on an OA
Furkus Xoud	Telum	Choking Tower	Shocking Grasp	FINALLY
Crane	Nadia	Choking Tower	Fireball	No seriously a crane
Xoud’s Assistant	Corrin	Choking Tower	Chainsaw	2 Crits in a row!
8 Legged Robot	Hardad	Choking Tower	EMP Pistol	Started combat with a crit to stun, never got an attack
Mishtu	Corrin	Choking Tower	Chainsaw	Smite Evil
Binox the Builder	Telum	Machine Caves	Shocking Grasp	Crit on the first round
Mi-Go	Telum	Fungal Caves	Scimitar	Crit as it disengaged to protect the caster!
Rainbow Mi-Go	Hardad	Fungal Caves	Pistol	Throwback kill
Gearsmen Battleguard	Hardad	Black Sovereign’s Palace	EMP Pistol	Stealth kills! apparently

What else?

There’s always more to look into! Give me all your ideas, questions, feedback, etc.

Iron Gods Report

Nat

2018-02-27

Session info

Killing blows

Trends by session

Trends by player character (PC)

Total kills per level

Total kills over time

Kills per session

Trends by location

Geographic Location

Specific sites

Who’s better where?

Trends by weapon

Overall

Per session

Creature Type

Creature Type by Geography

Location

Specific sites

Creature Types and PCs

Number of kills by PC

Proportion of kills by PC

Changes by PC over time

Perception of difficulty

Changes over time

Easier than expected

Harder than expected

Classification

Fun stuff

Alignment Chart

Boss list!

Notes

What else?