Speed up while loop in R - r

Question

As part of a project I made a smoother to smooth out missing data. I make use of the previous slope of the last data points to calculate new values. After calculated each new point I use this data to calculate a new value (and so on). Hence I used a while-loop to calculate each value (both from left to right as from right to left to eventually take a average of these 2 values). This scripts works fine!
Although I expect that I can significantly accelerate this with a function from the apply-family, I still want to use this while loop. The script is however really slow (3 days for ~ 2,500,000 data points). Do you have tips (for the current script) for me to change to speed things up?
#Loop from: bottom -> top
number_rows <- nrow(weight_id)
i <- nrow(weight_id)
while (i >= 1){
j = as.integer(weight_id[i,1])
prev1 <- temp[j+1,]$new_MAP_bottom
if(j<max(weight_id)){
previous_slope <- ifelse((temp[j+2,]$duration-temp[j+1,]$duration)>0,prev1-temp[j+2,]$new_MAP_bottom,0)
}else{
previous_slope <- 0
}
new_MAP <- round(prev1+((previous_slope-(factor*temp[j,]$steps))/(1+factor)), digit=2)
temp[j,]$new_MAP_bottom <- new_MAP
i <- i-1
}
#Loop from: top -> bottom
weight_factor <- 0
i <- 1
while (i <= nrow(weight_id)) {
j = as.integer(weight_id[i,1])
prev1 <- temp[j-1,]$new_MAP_top
if(j>2){
previous_slope <- ifelse((temp[j-1,]$duration-temp[j-2,]$duration)>0,prev1-temp[j-2,]$new_MAP_top,0)
}else{
previous_slope <- 0
}
new_MAP <- round(prev1+((previous_slope+(factor*temp[j,]$steps))/(1+factor)), digit=2)
temp[j,]$new_MAP_top <- new_MAP
#Take weighted average of two approaches (top -> bottom/bottom -> top)
if(weight_factor < 1){ weight_factor = temp[j,]$weight-1 }
weight_top <- weight_factor
weight_bottom <- temp[j,]$weight-weight_factor
if(weight_top>weight_bottom){ weight_top<-weight_top-1 }
if(weight_top<weight_bottom){ weight_bottom<-weight_bottom-1}
temp[j,]$MAP <- round(((new_MAP*weight_top)+(temp[j,]$new_MAP_bottom*weight_bottom))/(weight_top+weight_bottom),digit=0)
weight_factor <- weight_factor-1
i <- i+1
}

Answer 1

I did not read all of your code, especially without example data, but from the textual description, its only linear approximation: Please check, if the buildin functions approx and approxfun already do what you try to implement yourself, as these will be optimized more than you can with suitable effort.
par(mfrow=c(2,1))
example <- data.frame(x = 1:14,
y = c(3,4,5,NA, NA, NA, 6,7,8.1, 8.2, NA, 8.4, 8.5, NA))
plot(example)
f <- approxfun(example)
plot(example$x, f(example$x))
The apply family tends to give you shorter, more succinct code, but not necessarily much more speed then loops. If you are into speed, first check, if somebody else has already implemented, what you need, then try vectorization.
Edit:
The following runs in about a second on my computer. If this does something close enough to your own "linear smoother" so that you can replace yours with this, that is a speed increase of about 3 days.
n <- 2500000
example <- data.frame(x = 1:n,
y = sample(1:1000, n, replace = TRUE))
example$y[sample(1:n, n/5)] <- NA
print(Sys.time())
f <- approxfun(example)
mean(f(example$x))
print(Sys.time())