gistfile1.txt

#include <Rcpp.h>
using namespace Rcpp;

//' Find "detectable change points" in a numeric data frame
//'
//' If you have a huge data frame that you want to do a line-plot of, this
//' function can help down-sample the data to remove consecutive redundancies
//' that would be undetectable in a plot whose resolution is higher than
//' \code{thresh}. It's similar *in spirit* to something like
//' \code{x <- diff(as.matrix(df)); which(apply(x, 1, function(r) any(r>thresh)))},
//' but much faster, doesn't need additional memory, and catches slow drifts
//' that the aforecoded would miss.
//'
//' Our algorithm selects the first row in \code{df}, then progressively selects
//' the next row whose Manhattan distance is more than \code{thresh} away from
//' the previous selected row.  The first and last rows are always selected.
//'
//' @param df A numeric data frame.
//' @param thresh The threshold to screen by.
//' @return A numeric vector of indices into \code{df}.
//'
//' @export
//'
// [[Rcpp::export]]
std::vector<long> downsampleForPlot(DataFrame df, std::vector<double> thresh) {

  std::vector<double>::size_type m = df.length();
  if (m>1 && thresh.size() == 1) {
    thresh.resize(m, thresh[0]);
  } else if (m != thresh.size()) {
    Rcpp::stop("Vector of thresholds must be length 1 or the same length as the number of columns in 'df'");
  }

  std::vector<long> out(0);
  int n = df.nrows();
  if (n==0) return out;

  int last_i = 0;
  out.push_back(last_i + 1);


  NumericVector cols[m];
  for(size_t j=0; j<m; j++) {
    cols[j] = df[j];
  }

  for(int i=1; i<n; i++) {
    for(size_t j=0; j<m; j++) {
      if(fabs(cols[j][i]-cols[j][last_i])>thresh[j]) {
        out.push_back(i+1);
        last_i = i;
        break;
      }
    }
  }

  // Select the nth row if it wasn't already selected.
  if(last_i < n-1)
    out.push_back(n);

  return out;
}