Rcpp moving average - boundary error leads to fatal error - c++

Question

I use a rolling weighted moving average function whose code is provided below. It is coded in C++ via Rcpp.
This function works for most times series there is no loop issues or anything like this. I provided below a times series of length 2 that sometimes triggers the fatal error.
I could not find the reason of the error.
Thanks for your help! =)
Here is the R code :
# Install packages
sourceCpp("partialMA.cpp")
spencer_weights=c( -3, -6, -5, 3, 21, 46, 67, 0, 67, 46, 21, 3, -5, -6, -3)
spencer_ma <- function(x) roll_mean(x,spencer_weights)
x=c(11.026420323685528,0.25933761651337001)
spencer_ma(x) # works
for(i in 1:1000) spencer_ma(x) # triggers the fatal error
I include the C++ code of my roll_mean function below :
#include <Rcpp.h>
using namespace Rcpp;
// [[Rcpp::export]]
NumericVector roll_mean(const NumericVector& x,
const NumericVector& w) {
int n = x.size();
int w_size = w.size();
int size = (w_size - 1) / 2;
NumericVector res(n);
int i, ind_x, ind_w;
double w_sum = Rcpp::sum(w), tmp_wsum, tmp_xwsum, tmp_w;
// beginning
for (i = 0; i < size; i++) {
tmp_xwsum = tmp_wsum = 0;
for (ind_x = i + size, ind_w = w_size - 1; ind_x >= 0; ind_x--, ind_w--) {
tmp_w = w[ind_w];
tmp_wsum += tmp_w;
tmp_xwsum += x[ind_x] * tmp_w;
}
res[i] = tmp_xwsum / tmp_wsum;
}
// middle
int lim2 = n - size;
for (; i < lim2; i++) {
tmp_xwsum = 0;
for (ind_x = i - size, ind_w = 0; ind_w < w_size; ind_x++, ind_w++) {
tmp_xwsum += x[ind_x] * w[ind_w];
}
res[i] = tmp_xwsum / w_sum;
}
// end
for (; i < n; i++) {
tmp_xwsum = tmp_wsum = 0;
for (ind_x = i - size, ind_w = 0; ind_x < n; ind_x++, ind_w++) {
tmp_w = w[ind_w];
tmp_wsum += tmp_w;
tmp_xwsum += x[ind_x] * tmp_w;
}
res[i] = tmp_xwsum / tmp_wsum;
}
return res;
}

Answer 1

A Wild Index Out of Bounds Error Appeared!
You can pinpoint the issue by switching element accessors from [] to (). The latter has a built in bounds check, e.g. is index between 0 and n-1.
Running the code with the built-in check gives:
Error in roll_mean(x, spencer_weights) :
Index out of bounds: [index=7; extent=2].
So, the indices being used are greatly exceeding the length of the vector. Adding a trace statement indicates its the first loop that is wrong.
#include <Rcpp.h>
// [[Rcpp::export]]
NumericVector roll_mean(const NumericVector& x,
const NumericVector& w) {
int n = x.size();
int w_size = w.size();
int size = (w_size - 1) / 2;
Rcpp::Rcout << n << ", w_size: " << w_size << ", size: " << size << std::endl;
NumericVector res(n);
int i, ind_x, ind_w;
double w_sum = Rcpp::sum(w), tmp_wsum, tmp_xwsum, tmp_w;
// beginning
for (i = 0; i < size; i++) {
tmp_xwsum = tmp_wsum = 0;
// Fix this line
for (ind_x = i + size, ind_w = w_size - 1; ind_x >= 0; ind_x--, ind_w--) {
tmp_w = w(ind_w);
Rcpp::Rcout << "Loop at: " << ind_w << std::endl;
tmp_wsum += tmp_w;
tmp_xwsum += x(ind_x) * tmp_w;
}
res(i) = tmp_xwsum / tmp_wsum;
}
Rcpp::Rcout << "success" << std::endl;
return res;
}
And that's all folks!