temp.R

library(ISLR)

# prepare data example
df <- ISLR::Hitters[, c("Salary", "Years")]
row.names(df) <- NULL
df$Salary <- log(df$Salary)

# remove the observation with missing values in the response variable
df <- df[!is.na(df$Salary), ]
x <- df$Years
y <- df$Salary

# sort the variable to split
to_s <- sort(x)
# unique observation
to_s <- unique(to_s)

# create the mid-point as a cut-points candidate to split the space
s <- vector("double", length(to_s) - 1)
i <- 1
while (i < length(to_s)) {
  s[i] <- (to_s[i + 1] + to_s[i]) / 2
  i <- i + 1
}

# compute the loss function for each s candidate
output <- vector("double", length(s))
var <- x

for (i in 1:length(s)) {
  half_plane <- var > s[i]
  yhat1 <- mean(y[half_plane])
  yhat2 <- mean(y[!half_plane])
  e1 <- y[half_plane] - yhat1
  e2 <- y[!half_plane] - yhat2
  SRC1 <- sum(e1 ^ 2)
  SRC2 <- sum(e2 ^ 2)
  SRCT <- SRC1 + SRC2
  output[i] <- SRCT
  names(output)[i] <- s[i]
}


# plot the loss function; exist two possible values as minimum
p <- data.frame(x = as.double(names(output)), y = output,
                stringsAsFactors = FALSE)
plot(p$x, p$y, type = "l")
output[which.min(output)]