Detecting convergence
Great job iterating over the variables in the last exercise! But how many iterations are needed? When the imputed values don't change with the new iteration, we can stop.
You will now extend your code to compute the differences between the imputed variables in subsequent iterations. To do this, you will use the Mean Absolute Percentage Change function, defined for you as follows:
mapc <- function(a, b) {
mean(abs(b - a) / a, na.rm = TRUE)
}
mapc() outputs a single number that tells you how much b differs from a. You will use it to check how much the imputed variables change across iterations. Based on this, you will decide how many of them are needed!
The boolean masks missing_air_temp and missing_humidity are available for you, as is the hotdeck-initialized tao_imp data.
Este exercício faz parte do curso
Handling Missing Data with Imputations in R
Exercício interativo prático
Experimente este exercício completando este código de exemplo.
diff_air_temp <- ___
diff_humidity <- ___
for (i in 1:5) {
# Assign the outcome of the previous iteration (or initialization) to prev_iter
prev_iter <- ___
# Impute air_temp and humidity at originally missing locations
tao_imp$air_temp[missing_air_temp] <- NA
tao_imp <- impute_lm(tao_imp, air_temp ~ year + latitude + sea_surface_temp + humidity)
tao_imp$humidity[missing_humidity] <- NA
tao_imp <- impute_lm(tao_imp, humidity ~ year + latitude + sea_surface_temp + air_temp)
}