# Analyse births in Australia
library(tidyverse)
library(ggthemes)
library(forecast)
AU_births <- read_csv("data/BIRTHS_MONTH_OCCURRENCE_17102017230358770.csv")
# Exclude 2015 as data seems incomplete
AU_births <- filter(AU_births, !(TIME == 2015))
# Remove subtotals in rows
AU_births <- filter(AU_births, MONTH_OCCURENCE != "TOT")
# Make it a number as it was meant to be
AU_births$MONTH_OCCURENCE <- as.numeric(AU_births$MONTH_OCCURENCE)
# Add median by month to df
AU_births_medians <- AU_births %>%
  group_by(Region, MONTH_OCCURENCE) %>%
  summarise(median = median(Value))
AU_births <- merge(AU_births, AU_births_medians)

# Create a seasonal subseries plot
ggplot(filter(AU_births, Region == "Australia"), aes(x= Time, y = Value)) + 
  geom_line() +
  geom_line(aes(y = median), color = "red") +
  facet_grid( ~ MONTH_OCCURENCE) +
  labs(x = paste("\nFrom", min(AU_births$Time), "to",max(AU_births$Time)),
       caption = "Data: Australian Bureau of Statistics",
       y = "Births per month",
       title = "Births in Australia",
       subtitle = "Bottoms out in summer months; boomed from 2005/2007") +
  theme_economist() +
  theme(axis.text.x = element_blank()) 

# Let's check the seasonality
# Uses some base R graphics subtleties to right align the caption...
AU_births_ts <- as.ts(arrange(filter(AU_births, Region == "Australia"), Time, MONTH_OCCURENCE)$Value)
Acf(AU_births_ts, lag.max = 48, main = "")
    title(main = "ACF plot of births in Australia by month")
    title(sub = "Data: Australian Bureau of Statistics", adj = 1)

# The acf peaks at multiples of 12, showing a yearly seasonality

# Seasonality doesn't appear to increase over time
#plot(AU_births_ts)
#trend_AU_births <- ma(AU_births_ts, order = 12, centre = T) 
#plot(trend_AU_births)
#detrend_AU_births <- AU_births_ts - trend_AU_births

AU_births_decomposed <- decompose(ts(AU_births_ts, frequency = 12))
years <- rep(c(min(AU_births$TIME):max(AU_births$TIME)), each = 12)

# For ggplot, we prefer a long dataframe over a list of timeseries
AU_births_decomposed_df <- as.tibble(cbind(observation = c(1:length(years)),
                                 years,
                                 observed = as.vector(AU_births_ts), 
                                 seasonal = as.vector(AU_births_decomposed$seasonal),
                                 random = as.vector(AU_births_decomposed$random),
                                 trend = as.vector(AU_births_decomposed_trend)))
# Add a character vector only after the above cbind, to avoid everything becoming a chr or a factor
AU_births_decomposed_df <- mutate(AU_births_decomposed_df,
                                  seastrend = seasonal + trend,
                                  month = rep(c("01","02","03","04","05","06","07","08","09","10","11","12"), times = 19))  
AU_births_decomposed_df <- gather(AU_births_decomposed_df, key = component, value = value, -observation, -years, - month)

# Plot the various components of the observed births underneath each other
ggplot(filter(AU_births_decomposed_df, component != "seastrend"), aes(x = as.factor(observation), y = as.numeric(value), group  = 1)) +
  geom_line() +
  facet_grid(component ~ ., scales = "free_y") +
  # Hackish way of showing only the year
  scale_x_discrete(breaks = filter(AU_births_decomposed_df, month == "01")$observation,
                   labels = filter(AU_births_decomposed_df, month == "01")$years) +
  theme_fivethirtyeight() +
  theme(axis.text.x = element_text(angle = 45)) +
  labs(title = "Births in Australia",
       subtitle = "Seasonally decomposed",
       caption = "Data: Australian Bureau of Statistics",
       x = "",
       y = "")

# Let's look at everything in one plot instead
ggplot(AU_births_decomposed_df, aes(x = yearmonth, y = as.numeric(value), group  = 1)) +
  geom_line(data = dplyr::filter(AU_births_decomposed_df, component =="trend"), color = "orange", size = 2) +
  geom_line(data = dplyr::filter(AU_births_decomposed_df, component =="observed"), color = "grey50") +
  geom_line(data = dplyr::filter(AU_births_decomposed_df, component =="seastrend"), color = "black") +
  scale_x_discrete(breaks = filter(AU_births_decomposed_df, month == "01")$yearmonth,
                   labels = filter(AU_births_decomposed_df, month == "01")$years) +
  theme_economist_white() +
  theme(axis.text.x = element_text(angle = 45)) +
  labs(title = "Births in Australia",
       subtitle = "Seasonally decomposed",
       caption = "Data: Australian Bureau of Statistics",
       x = "",
       y = "")

# TODO Horizon plot