platano<- ts(platano$produccion,start = c(2007),frequency = 12)

plot.ts(platano);title("Produccion de babano en el Peru")

autoplot(platano)+theme_minimal()+labs(title = "Produccion de babano en el Peru",
                                       x="tiempo")+geom_line(color="blue")

ts_decompose(platano)

# En este paso obtenemos los primeros 12 valores de la seria en donde tomamos l factor estacional. 
# Son 12 debdio que en el componente estacional siemple se repite un patron.

estacional <- decompose(platano)$seasonal
factor<- estacional[1:12]

# Ahora se crea un dataframe en donde se toma dos culumnas una por los meses y la otra de los valores de 
# los factores

base_e <- data.frame("n"=1:12,"com"=factor)

# Ahora se crea el grafico 

base_e %>% ggplot(aes(n,com))+geom_col(fill=ifelse(factor>0,"blue","red"))+
  theme_minimal()+labs(title = "comportamiento estacional",
                       subtitle = "Producción de platano",
                       x="Meses",y=" ")+geom_hline(yintercept = 0,
                                                   color="black",size=0.8)+
  geom_line()+geom_point(color=ifelse(factor>0,"blue","red"),size=2)

platano <- as.xts(platano)  
tiempo<- index(platano)
base <- data.frame("Fecha"=tiempo,"Produccion"=platano,anio=year(tiempo),
                   mes=month(tiempo))

base %>% head(5) %>% kbl() %>% kable_material_dark() %>%  
  column_spec(column = 1, width = "1cm", background = alpha("blue",0.3)) 

base %>% 
  mutate(
    weekday = MONTH(Fecha, label = TRUE, abbr = FALSE), 
    weekday = weekday %>% as_factor() 
  ) %>% 
  ggplot(aes(x=weekday, y=Produccion)) +
  geom_jitter(aes(color=weekday),
              alpha = 0.5,
              size=0.5,
              show.legend = FALSE) + 
  geom_half_violin(aes(fill=weekday),
                   side = "l",
                   alpha = 0.45,
                   show.legend = FALSE,
                   trim = FALSE) +
  geom_half_boxplot(aes(fill = weekday),
                    side = "r",
                    outlier.size = 3,
                    outlier.color = "red",
                    width = 0.4, 
                    alpha = 0.45,
                    show.legend = FALSE) + 
  stat_summary(fun = median, geom="line")  +
  theme_bw() 

base %>% as_tibble()-> ata
s<- ata %>% plot_seasonal_diagnostics(Fecha,Produccion,.interactive = T)

s

modelo1 <- auto.arima(train)
summary(modelo1)

modelo2 <- nnetar(train)
summary(modelo2)

df <- base %>% select(Fecha,Produccion)
names(df)<- c("ds","y")
modelo3 <- prophet(df)

pred1 <- forecast(modelo1,12)
pred2 <- forecast(modelo2,12)
futuro <- make_future_dataframe(modelo3, periods = 12)
pred3 <- predict(modelo3, futuro)
proy <- pred3[["yhat"]] %>% tail(12)

data_pred<- data.frame(n=1:12,sarima=pred1$mean,nnet=pred2$mean,prophet=proy,test=test)

data_pred %>% ggplot(aes(n,test)) + geom_line(aes(color="test")) + 
  geom_line(mapping = aes(x=n,y=sarima,color="sarima"))+ geom_line(mapping = aes(x=n,y=nnet,color="nnet"))+
   geom_line(mapping = aes(x=n,y=prophet,color="prophet"))

p <- c(RMSE(test,pred1$mean),MAPE(test,pred1$mean),RMSLE(test,pred1$mean))
p1 <- c(RMSE(test,pred2$mean),MAPE(test,pred2$mean),RMSLE(test,pred2$mean))
p2 <- c(RMSE(test,proy),MAPE(test,proy),RMSLE(test,proy))
dac <- cbind(p,p1,p2) %>% as.data.frame()

dac %>% kbl() %>% kable_material_dark() %>%  
  row_spec(2,  background = alpha("blue",0.3)) 

modelo2 <- nnetar(platano)
prednnet <- forecast(modelo2,12)
autoplot(prednnet)