These are chat archives for symengine/symengine

13th
Dec 2015
Ondřej Čertík
@certik
Dec 13 2015 05:17
@rwst thanks for the timings. They look pretty encouraging. It would be also interesting to benchmark ginac directly in C++, because I thought the Pynac wrappers cause some kind of overhead, i.e. that it is calling Python for the arithmetic. As to Flint, that looks like it is caused by the fact that Piranha is good at multivariate stuff, but not as good at single variate stuff. We can later provide a specialized version for univariate if needed.
How do you propose to override the series in SymPy? What I was thinking is that the series method would be able to accept a method parameter, which you can use to choose between the current SymPy's implementation, the coming rs_series, and if symengine is installed, the symengine version. For now I would make symengine optional.
@CodeMaxx there are lots of features still missing in SymEngine, as @rwst mentioned. You can pick something that interests you and implement it in C++.
Ralf Stephan
@rwst
Dec 13 2015 07:01
@certik The Pynac timings are with GMP bigints and rationals which is in the newest Sage dev version (together with C++11). To override series in SymPy using method= is a natural idea. ATM I simply replace series and call SymPy's version for multivar expressions but I planned to change it today to wrap a try...except around it and call SymPy when SymEngine throws an exception. Now I could combine everything.