optimized equi-width histogram

* Now, histogram is build based on min, max, and number of buckets. Thus, we optimized its memory usage and improved runtime. We do not need to store bucket border values anymore and bucket search is now O(1) and no need for binary search.
* Comparing floating point numbers still happen but only comparing with min and max values from different histogram object. To address the known issues in comparing float/double values, adaptive approach is integrated to dynamically select the epsilon. This is expected to work well with large values as well as numbers close to zero.
* Added support for new data types such as float, i8, and ui8.
commit_hash:cc76490319c2b3708e6c8d9bc299ba415a9b2127

0 files changed, 0 insertions, 0 deletions