Releases: apache/datasketches-python
Releases · apache/datasketches-python
5.2.0
5.1.1
v5.1.0
This is a minor release version with several new features and one change to wheels:
- Convenience wheels in pypi are now built using numpy 2.x, but should still work if using numpy 1.x
- New sketch: The python library now supports the DataSketches implementation of t-digest
- The tuple sketch supports a .filter() method which applies a predicate to the summaries and returns a compact sketch of items matching the predicate
v5.0.2
5.0.1
This is also a patch release that fixes an issue with wheels for Apple Silicon Macs. As a rapid-follow match to a major release, the release notes describing the significant changes are:
- Vastly improved API documentation via sphinx
- More pythonic API:
- copy instead of C++-style copy constructors
- str takes no arguments in all cases
- Moved from pybind11 to nanobind, which also requires C++17 internally
- Added new exact proportional to size sampling sketch, ebpps_sketch
- Fixed slowness with theta intersection
- Fixed incompatibility with Java's serialized empty frequent items sketch
- Removed default Kernel for density sketch (GaussianKernel is still provided)
- Removed extra (largely transparent to users) wrappers around density_sketch and tuple_sketch
- Added NOMINSIZE to nanobind config, increasing performance at the cost of a slightly larger wheel
5.0.0
This is also a major version release with several significant changes:
- Vastly improved API documentation via sphinx
- More pythonic API:
__copy__instead of C++-style copy constructors__str__takes no arguments in all cases
- Moved from pybind11 to nanobind, which also requires C++17 internally
- Added new exact proportional to size sampling sketch, ebpps_sketch
- Fixed slowness with theta intersection
- Fixed incompatibility with Java's serialized empty frequent items sketch
- Removed default Kernel for density sketch (GaussianKernel is still provided)
- Removed extra (largely transparent to users) wrappers around density_sketch and tuple_sketch
- Added NOMINSIZE to nanobind config, increasing performance at the cost of a slightly larger wheel
Known issue:
- Wheels intended for Apple Silicon Macs are mistakenly compiled for x86_64 architecture (will be fixed in forthcoming 5.0.1)