pyDatalog 0.16.0rc1

[Pierre Carbonnelle]

Dear all,

I'm pleased to announce the availability of pyDatalog 0.16.0rc1.

The resolution engine has been re-engineered to dramatically improve performance of several types of algorithms.  In previous versions, a negated literal in the body of a clause was resolved as a separate goal : as a consequence, intermediate results found during its resolution could not be re-used : so the same facts possibly had to be established multiple times in the same query.  Similarly, aggregate clauses were resolved in isolation.  As a consequence, reports computed in pyDatalog, such as the PYPL index, were very slow.

Finally, the new resolution engine is now able to switch from depth-first to breadth-first exploration of the computation tree as appropriate.  With this, pyDatalog can be used both for rule-based expert system (depth-first) and for graph analysis (breadth-first).  See issue #7.  Thanks to this, the hashtag problem can now be solved under 7 seconds.

Because of the significant re-write, this is a release candidate.  Please let me know if you have any issue with it.

Best regards,

Pierre Carbonnelle