ajuste_lm_serie_temporal.R

# Codigo inspirado neste post:
# https://www.business-science.io/code-tools/2017/10/24/demo_week_timetk.html
# Testando se o ajuste funciona para a serie AirPassenger
# Nao sei ate que ponto este metodo esta correto... Vamos estudar!

library(tsibble)
library(lubridate)
library(dplyr)

library(timetk)     # Toolkit for working with time series in R
library(tidyquant)  # Loads tidyverse, financial pkgs, used to get data

# Obtendo serie:
serie <- AirPassengers

# Serie no formato "tidy"
serie <- 
  serie %>% 
  as_tsibble() %>% 
  mutate(date = as.Date(index)) %>% 
  as_tibble() %>% 
  select(-index)

# Removendo ultimos 6 valores para teste:
actuals_tbl <- tail(serie, 6)
serie       <- serie %>% .[1:(nrow(.)-6),]

# Plot Serie
serie %>%
  ggplot(aes(date, value)) +
  geom_line(col = palette_light()[1]) +
  geom_point(col = palette_light()[1]) +
  geom_ma(ma_fun = SMA, n = 12, size = 1) +
  theme_tq() +
  scale_x_date(date_breaks = "1 year", date_labels = "%Y") +
  labs(title = "Airpassenger")
 
# Obtendo variaveis de calendario:
serie <- 
  serie %>%
  tk_augment_timeseries_signature() 

# linear regression model used, but can use any model
fit_lm <- lm(value ~ ., data = select(serie, -one_of(c("date", "diff"))))

# Resultado do modelo:
summary(fit_lm)

# Retrieves the timestamp information
serie_idx <-
  serie %>%
  tk_index()

tail(serie_idx)

# Make future index
future_idx <- 
  serie_idx %>%
  tk_make_future_timeseries(n_future = 6)

future_idx

new_data_tbl <- 
  future_idx %>%
  tk_get_timeseries_signature()

new_data_tbl

# Make predictions
pred <- predict(fit_lm, newdata = select(new_data_tbl, -one_of(c("index", "diff"))))

predictions_tbl <- tibble(
  date  = future_idx,
  value = pred
)

predictions_tbl

actuals_tbl

# Plot da previsao + valores reais:
serie %>%
  ggplot(aes(x = date, y = value)) +
  # Training data
  geom_line(color = palette_light()[[1]]) +
  geom_point(color = palette_light()[[1]]) +
  # Predictions
  geom_line(aes(y = value), color = palette_light()[[2]], data = predictions_tbl) +
  geom_point(aes(y = value), color = palette_light()[[2]], data = predictions_tbl) +
  # Actuals
  geom_line(color = palette_light()[[1]], data = actuals_tbl) +
  geom_point(color = palette_light()[[1]], data = actuals_tbl) +
  # Aesthetics
  theme_tq() +
  labs(title = "AirPassenger: Time Series Machine Learning",
       subtitle = "Using basic multivariate linear regression can yield accurate results")