predict_10000pkgs_eoda.R

# get data ----------------------------------------------------------------
library(Ecdat)
data("CRANpackages")
df <- rbind(CRANpackages, 
            data.frame(Version = c("3.2", "3.3") ,
                       Date = as.Date(c("2016-04-27", "2016-12-22")), 
                       Packages = c(8329, 9687),
                       Source = c("Andrie de Vries", "Eduard Szöcs")
            )
)


# convert date to numeric
df$date_d <-  as.numeric(df$Date)



# plot data ---------------------------------------------------------------

ggplot(df, aes(x = Date, y = Packages)) +
  geom_point()


ggplot(df, aes(x = date_d, y = log(Packages))) +
  geom_point()

# Model data --------------------------------------------------------------

library(mgcv)
mod_gam <- gam(Packages ~ s(date_d), data = df, method = 'REML')
gam.check(mod_gam)


# predict data
pdat <- data.frame(date_d = seq(min(df$date_d), 17300, 1))
pdat$Date <- as.Date(pdat$date_d, origin = '1970-01-01')
pdat$pred <- predict(mod_gam, newdata = pdat)

# check predictions for 10000
pdat[pdat$pred > 9997 & pdat$pred < 10003, ]


ggplot() +
  geom_point(data = df, aes(x = Date, y = Packages)) +
  geom_line(data = pdat, aes(x = Date, y = pred)) +
  geom_hline(aes(yintercept = 10000), linetype = 'dotted') +
  geom_vline(aes(xintercept = 17248), linetype = 'dotted')