Conditioned Moving Max

Question

I'm a relative R noob.

I've got a big dataset that looks something like this:

         Tempadjvolt      newmass     rgdeltas
2794       498.5777  0.5355647187   0.00000000
2795       499.7577  0.5355647187   0.00000000
2796       500.7877  0.3415104788  -2.87487763
2797       502.1177  0.4312854788  -1.54487763
2798       500.3877  0.5355647187   0.00000000
2799       502.5377  0.4596354788  -1.12487763
2800       507.6877  0.8072604788   4.02512237
2801       505.2577  0.6432354788   1.59512237
2802       505.7977  0.6796854788   2.13512237
2803       517.8877  1.4957604788  14.22512237
2804       502.2477  0.4400604788  -1.41487763
2805       507.3677  0.7856604788   3.70512237
2806       519.7277  1.6199604788  16.06512237
2807       528.9377  2.2416354788  25.27512237
2808       520.2677  1.6564104788  16.60512237
2809       519.3877  0.5355647187   0.00000000
2810       526.5677  2.0816604788  22.90512237
2811       519.5377  0.5355647187   0.00000000
2812       526.9277  2.1059604788  23.26512237
2813       529.9877  2.3125104788  26.32512237
2814       514.4077  1.2608604788  10.74512237
2815       518.3777  1.5288354788  14.71512237

I'm trying to identify negative rgdeltas values [for example, row 2804] and then 'look' 7 positions behind and ahead to find the highest Tempadjvolt and set row 2804's tempadjvolt to that local max.

The frame is ~4000 rows long, of which ~515 are negative values. I tried a couple for loops that sorta worked... but also spit out a bunch of NAs -- which makes me think they were poorly/improperly constructed.

Any assistance would be greatly appreciated.

As was pointed out in the comments, the original post was unclear. I'm not concerned about consecutive negative rgdeltas values. For negative values within 7 of the front and end of the frame, ideally the loop would look as many positions forward and back before the beginning/end. Less concerned with that at this point.

A little more background: This is part of a signal processing program originally written in C# that I'm attempting to move to R to augment some more facile batch processing of a large number of files output from an environmental monitor. I didn't write the original code and this is only one small component of a much larger set of stuff going on.

I appreciate the help. Thanks!

Answer 1

1) Zero Fill. Assuming that the data frame is called DF we use rollapply in the zoo package to apply function, f, to a moving window of width 15:

library(zoo)
# columns of DF are (1) Tempadjvolt, (2) newmass and (3) rgdeltas
f <- function(x) if (x[8, 3] < 0) max(x[, 1]) else x[8, 1]
DF[[1]] <- rollapply(DF, 15, f, fill = 0, by.column = FALSE)

In the above we filled the points near the beginning and end with zeros since it seems the precise way of dealing with this is not so important but we could have used some other fill value.

2) Leave end values. Another possibility is to only process the points not near the ends:

DF[seq(8, nrow(DF)-7), 1] <- rollapply(DF, 15, f, by.column = FALSE)

3) Partials. or we could have used partial = TRUE and then take the max of the partial values near the ends like this:

f2 <- function(x) { 
      # Columns of DF2 are (1) Tempadjvolt, (2) newmass, (3) rgdeltas and (4) seq.
      # Condition is TRUE if passed a partial x near the beginning.
      # k is row index of current row in x. Normally 8 but near start it varies.
      k <- if (x[1, 4] == 1) nrow(x) - 7 else 8
      if (x[k,  3] < 0) max(x[, 1]) else x[k, 1]
}
DF2 <- cbind(DF, seq = 1:nrow(DF))
DF[[1]] <- rollapply(DF2, 15, f2, partial = TRUE, by.column = FALSE)

Answer 2

Assume its name is dat:

negidxs <- as.numeric( rownames(dat)[ dat[[3]] < 0 ] )
 for ( i in negidxs ){ 
      dat[as.character(i), "Tempadjvolt"] <- 
            max(dat[rownames(dat) %in% (i-7):(i+7), "Tempadjvolt"], na.rm=TRUE) }
 dat
  #----------------------------------#
     Tempadjvolt   newmass  rgdeltas
2794    498.5777 0.5355647  0.000000
2795    499.7577 0.5355647  0.000000
2796    517.8877 0.3415105 -2.874878
2797    517.8877 0.4312855 -1.544878
2798    500.3877 0.5355647  0.000000
2799    519.7277 0.4596355 -1.124878
2800    507.6877 0.8072605  4.025122
2801    505.2577 0.6432355  1.595122
2802    505.7977 0.6796855  2.135122
#snipped-----