Commits
Wes McKinney authored 155bf076204
ARROW-1854: [Python] Use pickle to serialize numpy arrays of objects. **Just posting this for discussion.** See the preceding discussion on https://issues.apache.org/jira/browse/ARROW-1854. I think the ideal way to solve this would actually be to improve our handling of lists, which should be possible given that pickle seems to outperform us by 6x according to the benchmarks in https://issues.apache.org/jira/browse/ARROW-1854. Note that the implementation in this PR will not handle numpy arrays of user-defined classes because it will not fall back to cloudpickle when needed. cc @pcmoritz @wesm Author: Wes McKinney <wes.mckinney@twosigma.com> Author: Robert Nishihara <robertnishihara@gmail.com> Closes #1360 from robertnishihara/numpyobject and squashes the following commits: c37a0a08 [Wes McKinney] Fix flake 51915032 [Wes McKinney] Fix post rebase 43f2c805 [Wes McKinney] Add SerializationContext.clone method. Add pandas_serialization_context member that uses pickle for NumPy arrays with unsupported tensor types c9440231 [Wes McKinney] Use pickle.HIGHEST_PROTOCOL, convert to Buffer then memoryview for more memory-efficient transport cf719c3f [Robert Nishihara] Use pickle to serialize numpy arrays of objects.