Improving boxplot performance

Hi,

I’m trying to use dc.js to generate dynamic boxplots, but was having some performance issues with larger datasets (~50k float entries). I narrowed the issue down to this line under the reduce().reduceRemove function (taken from example):

p.splice(p.indexOf(v.Speed), 1);

I was thinking that if the p returned was pre-sorted (I’d imagine that the boxplot function already sorts the array to calculate quartiles) so that a more efficient binary search tree can be used to find the index for the splice. Is this something that can be implemented?

Thanks

[mrmin123]

Hi. Not sure what you mean by profiling this. If you mean benchmarking, no real benchmarks, but there were noticeable differences when I used d3.bisectLeft as you suggested. Huge improvements when combined with underscore.js’s indexOf function flagged to use binary search:

speedArrayGroup = experimentDimension.group().reduce(
        function(p,v) {
          p.splice(d3.bisectLeft(p, v.Speed), 0, v.Speed);
          return p;
        },
        function(p,v) {
          p.splice(_.indexOf(p, v.Speed, true), 1);
          return p;
        },
        function() {
          return [];
        }
      );

If you feel that this can be further improved, please let me know.

Thanks!

[gordonwoodhull]

Moving the pre-sorting option to #1527. Fixing the examples now.