ipi.df <- read_csv('data/ipi.csv')

#----------------------------------------------------------------------------
# Exercise 1. Transform ipi.df from:
#
# date      ipi
# <chr>   <dbl>
# 2017M11 104
# to:
#
# date         ipi
# <date>     <dbl>
# 2017-11-01 104
#----------------------------------------------------------------------------

# Sol:
ipi.df %<>%
  separate(date, into = c("year", "month"), sep = "M") %>%
  mutate(date = as.Date(paste(year, month, 01, sep = "-"))) %>%
  select(date, ipi)

#----------------------------------------------------------------------------
# Exercise 2. Plot the data as a time series using ggplot2
#----------------------------------------------------------------------------

ipi.df %>%
  ggplot(aes(x = date, y = ipi)) +
  geom_line(size = 1, color = palette_light()[[1]]) +
  geom_smooth(method = "loess") +
  labs(title = "Indice De Produccion Industrial De Cantabria", x = "", y = "IPI") +
  scale_y_continuous() +
  scale_x_date(date_breaks = "1 year", date_labels = "%Y") +
  theme_tq()

#----------------------------------------------------------------------------
# Exercise 3. Using ggplot, plot ipi.df IPI series with the the
# short_term_mean and the long_term_mean
#----------------------------------------------------------------------------

ipi.df %>%
  mutate(short_mean = short_term_mean(ipi),
         long_term_mean = long_term_mean(ipi)) %>%
  gather(key = "indice", value ='val', -date) %>%
  ggplot(aes(x = date, y = val, color=indice)) +
  geom_line(size = .5, color = palette_light()[[1]]) +
  geom_smooth(method = "loess") +
  labs(title = "Indice De Produccion Industrial De Cantabria", x = "", y = "IPI") +
  scale_y_continuous() +
  scale_x_date(date_breaks = "1 year", date_labels = "%Y") +
  theme_tq()

#----------------------------------------------------------------------------
# Exercise 4. Fit a Holt-Winters leaving all parameter in blank.
# Get the proposed α, β and γ.
# Check the final sum of squared errors achieved in optimizing
#----------------------------------------------------------------------------
fit_hw <- HoltWinters(ipi_train)
round(fit_hw$alpha, 2)
round(fit_hw$beta, 2)
round(fit_hw$gamma, 2)
plot(fit_hw)
fit_hw$SSE