Data Frame example:

Country= c('Angola', 'Angola', 'Angola', 'Angola', 'Angola', 'Angola',  'Algeria', 'Algeria', 'Algeria', 'Algeria', 'Algeria', 'Algeria')
Year= c( 2000, 2000, 2001, 2001, 2002,2002, 2000, 2000, 2001, 2001, 2002,2002)
Species= c( 'Goats', 'Sheep ', 'Goats', 'Sheep',  'Goats', 'Sheep',  'Goats', 'Sheep',  'Goats', 'Sheep',  'Goats', 'Sheep') 
Pop= c(20, 30, 22, 34, 18, 35, 24, 26, 40, 30 , 23, 43)

data <- data.frame(Country, Year, Species, Pop)

The loop function to predict animal population for the next 10 years per country and per specie :

# Create a loop over each country
for (country in unique(data$Country)) {
  # Create a loop over each species
  for (species in unique(data$Species)) {
    # Filter the dataframe by country and species
    temp <- subset(data, Country == country & Species == species)
    # Fit the ARIMA model
    model <- auto.arima(temp$Pop)
    # Make the predictions
    pred <- predict(model, n.ahead = 10)
    # Store the predictions in a dataframe
    if (exists("newdata")) {
      newdata <- rbind(newdata, data.frame(
        Country = country, 
        Year = seq(max(temp$Year) 1, max(temp$Year) 10), 
        Species = species, 
        Predicted_Pop = pred)
      )
    } else {
      newdata <- data.frame(
        Country = country, 
        Year = seq(max(temp$Year) 1, max(temp$Year) 10),    
        Species = species, 
        Predicted_Pop = pred
      )
    }
  }
}

The code works on a singular country but does not work when looping , I get the following error:

Error in [<-.ts(*tmp*, ri, value = c(7990484, 7990484, 7990484, 7990484, : only replacement of elements is allowed

I understand the arima model is creating two lists (pred and se) time series object and rbind is the one not able to combine it as a dataframe? Any pointers will be appreciated. I tried replace() function but still got the same error. Thanks!

I expect the loop to produce a list of animal population for each country and each species for future 10 years.

SOLVED Thanks to @Parfait suggestion I used his approach within my initial code:

        library(forecast) 
        results<- data.frame()
        
        # Create a loop over each country
        for (country in unique(data$Country)) {
           for (species in unique(data$Species)) {
        
        # Filter the dataframe by country and species
        temp <- subset(FAO.DB, Country == country & Species == species)
        
        # Check if the dataframe is empty
        if(nrow(temp) > 1){
        # Fit the ARIMA model
        model <- auto.arima(temp$Pop)
        
        # Make the predictions
        pred <- forecast(model, n.ahead = 10)
    
        # Store the predictions in a dataframe
        newdata <- data.frame( Country = country, 
                             Year = seq(max(temp$Year) 1, 
                                    max(temp$Year) 10), 
                             Species = species, 
                             Predicted_Pop = pred)
    
        # Store the predictions in a dataframe
        
        results <- rbind(results, newdata)
    
               }
             }
         }

I have also changed function predict() by the function forecast()

CodePudding user response：

Issue likely is due to an empty temp when slicing data frame by unique values of Country and Species. Instead of nested for loops, consider generializing your process in a single function, then split your data frame into those groups with by or split, and then process each split through your defined method. This will have you working with lists of data frames that you can then rbind once at the end.

# User-defined method
predict_population <- function(temp) {
  if (NROW(temp) > 1) {
    # Fit the ARIMA model
    model <- auto.arima(temp$Pop)
    # Make the predictions
    pred <- predict(model, n.ahead = 10)
  
    # Store the predictions in a dataframe
    newdata <- data.frame(
      Country = temp$Country[1], 
      Year = seq(max(temp$Year) 1, max(temp$Year) 10), 
      Species = temp$Species[1], 
      Predicted_Pop = pred
    )
  }
}

# APPROACH 1:
# Filter the dataframe by country and species with `by`  
df_list <- by(data, data[c("Country", "Species")], predict_population)

# APPROACH 2:
# Filter the dataframe by country and species with `split`
# Run method with `lapply`
df_list <- split(data, data[c("Country", "Species")]) |> lapply(predict_population)

# RBIND ALL DFs TO MASTER DATA FRAME
new_data <- do.call(rbind, unname(df_list))