Introduction
I was working on an application to show 3D paths using Direct3D and ran into a video card limit in the number of elements in the path. To solve the problem, I wrote some simple code to sample the data I was generating without just clipping the data off at a fixed point. While I needed it for this special case, the problem is a general one that arises in stretching or shrinking data such as images or changing the sample rate of audio files.
I searched for some code that would accomplish that and the closest I could come up with was writing a custom predicate for the boost filter iterator. While writing the predicate, I realized there was a more general solution that would support not just sampling the data set but stretching it as well.
In order to provide the most flexible and easy to use interface I provided 3 ways that the sample algorithm can be used. Through an input iterator, an output iterator or a predicate that can be used by the boost filter iterator or the STL remove_if algorithm.
Using the code
Using the iterator is relatively simple. Per the STL convention, I've provided a pair of template functions that allow for easy creation of both input and output iterator forms. Here is a simple sample that shows the use of both forms input and output.
std::vector list;
for(int i=0; i<7; ++i)
list.push_back(i);
std::cout << "Original sequence:" << std::endl;
std::copy(list.begin(), list.end(), std::ostream_iterator(std::cout, " "));
std::cout << std::endl;
std::cout << "Input sampled/stretched at various intervals:" << std::endl;
for(size_t dest_size = 1; dest_size<=list.size()*2; ++dest_size)
{
std::copy(sample::sample(list.begin(), dest_size, list.size()),
sample::sample(list.end(), dest_size, list.size()),
std::ostream_iterator(std::cout, " "));
std::cout << std::endl;
};
std::cout << "Output sampled/stretched at various intervals:" << std::endl;
for(size_t dest_size = 1; dest_size<=list.size()*2; ++dest_size)
{
std::copy(list.begin(), list.end(), sample::sample_output(
std::ostream_iterator(std::cout, " "), dest_size, list.size()));
std::cout << std::endl;
};
The biggest limitation of these objects is that you must know the size of both the input and output dataset. Also, a simple sampling method is used that only takes the location of the element in the data set into account. It doesn't look at the quality of the sample point to drop. For example an optimal solution for sampling a 3D curve would take the curvature at each point into account when deciding to drop a data point.
Points of Interest
I suspected that moving the ownership of the actual base iterator object out of the functor object into the wrapping iterator objects would provide a greater flexibility and ease of use. I'm working on some more advanced sample functors right now that will use averaging/linear interpolation as well as spline interpolation, and I think these will be simplified by owning the iterator since they will make it easier for the algorithm to read/write the interpolated data without having excessive access to the iterator class internal objects. I hope to post the updated functors in the near future and will include any updates to these iterators that are required to make it work.
