MachLearn02

hod2 <- count(deaths, c("hod", "cod"))
hod2 <- subset(hod2, !is.na(hod))
hod2 <- join(hod2, codes, by = "cod")
hod2 <- ddply(hod2, "cod", transform, prop = freq / sum(freq))
overall <- ddply(hod2, "hod", summarise, freq_all = sum(freq))
overall <- transform(overall, prop_all = freq_all / sum(freq_all))
hod2 <- join(hod2, overall, by = "hod")
devi <- ddply(hod2, "cod", summarise, n = sum(freq),
dist = mean((prop - prop_all)^2))
devi <- subset(devi, n > 50)

ggplot(data = devi, aes(x = n, y = dist)) + geom_point()
last_plot() + scale_x_log10() + scale_y_log10() +
geom_smooth(method = "rlm", se = FALSE)
ggsave("n-dist-resid.pdf", width = 6, height = 6)

devi$resid <- resid(rlm(log(dist) ~ log(n), data = devi))
unusual <- subset(devi, resid > 1.5)
hod_unusual <- match_df(hod2, unusual)
ggplot(hod_unusual_big, aes(hod, prop)) +
  geom_line(aes(y = prop_all),
  data = overall, colour = "grey50") +
  geom_line() +
  facet_wrap(~ disease, ncol = 3)

ggsave("unusual-big.pdf", width = 8, height = 6)
last_plot() %+% hod_unusual_sml
ggsave("unusual-sml.pdf", width = 8, height = 4)