impPCA.R
In VIM: Visualization and Imputation of Missing Values

## ----include = FALSE----------------------------------------------------------
knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>",
  fig.width = 7,
  fig.height=4,
  fig.align = "center"
)

## ----message=FALSE------------------------------------------------------------
library(VIM)

data(iris)
df <- iris[1:30,c(1,4)]
colnames(df) <- c("S.Length", "P.Width") # select two numerical variables
df_na <- df

# randomly produce some missing values in the column P.Width
set.seed(1)
nbr_missing <- 10
index_na <- sample(nrow(df_na),size = nbr_missing,replace = T)
index_na <- index_na[!duplicated(index_na)]
df_na[index_na,2] <- NA

w <- is.na(df_na$`P.Width`)

aggr(df_na)


## ----message=FALSE, results='hide'--------------------------------------------
imputed <- impPCA(df_na, method = "mcd", boot=TRUE)[[1]]
aggr(imputed)


## ----message=FALSE, fig.height=5----------------------------------------------
# create plot
plot(`P.Width` ~ `S.Length`, data = df, type = "n", ylab = "P.Width", xlab="S.Length")
mtext(text = "impPCA robust", side = 3)
points(df$`S.Length`[!w], df$`P.Width`[!w])
points(df$`S.Length`[w], df$`P.Width`[w], col = "grey", pch = 17)
points(imputed$`S.Length`[w], imputed$`P.Width`[w], col = "red", pch = 20, cex = 1.4)
segments(x0 = df$`S.Length`[w], x1 = imputed$`S.Length`[w], y0 = df$`P.Width`[w],
y1 = imputed$`P.Width`[w], lty = 2, col = "grey")
legend("topleft", legend = c("non-missings", "set to missing", "imputed values"),
pch = c(1,17,20), col = c("black","grey","red"), cex = 0.7)
mape <- round(100* 1/sum(is.na(df_na$`P.Width`)) * sum(abs((df$`P.Width` -
imputed$`P.Width`) / df$`P.Width`)), 2)
s2 <- var(df$`P.Width`)
nrmse <- round(sqrt(1/sum(is.na(df_na$`P.Width`)) * sum(abs((df$`P.Width` -
imputed$`P.Width`) / s2))), 2)
text(x = 5.6, y = 0.16, labels = paste("MAPE =", mape))
text(x = 5.6, y = 0.12, labels = paste("NRMSE =", nrmse))