[GSoC 2018] Automatic differentiation using tangent

[Navjot Kukreja]

Hello everyone

I am Navjot Kukreja, a PhD student at Imperial College London. I would like to propose a GSoC project to enable automatic differentiation of Sympy constructs using tangent (https://github.com/google/tangent). Tangent does source-to-source automatic differentiation directly on python code. The ability to perform automatic differentiation on python code that contains sympy constructs would enable/simplify the use of sympy in defining various optimization problems, which is a very broad class of problems in HPC/Science/Engineering. It also makes it easier to use sympy in the construction of DSL compilers.

Has this been considered before? Is there someone in the community that has tried this or needs this?

Thank you

Navjot Kukreja

https://github.com/navjotk

PS: I am one of the developers of Devito (https://github.com/opesci/devito), a sympy-based DSL for the rapid development of high-performance finite-difference simulations. The primary use of devito is as part of PDE-constrained optimization problems where the problem definition, including the objective function, governing PDE, interpolators etc. are provided as a combination of sympy expressions and pure python.

[Aaron Meurer]

I haven't heard of tangent before.

Can you clarify more what the project would do? Wouldn't it be enough
to just run sympy.python() on an expression, then pass that to
tangent?

Aaron Meurer

> --
> You received this message because you are subscribed to the Google Groups
> "sympy" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to sympy+un...@googlegroups.com.
> To post to this group, send email to sy...@googlegroups.com.
> Visit this group at https://groups.google.com/group/sympy.
> To view this discussion on the web visit
> https://groups.google.com/d/msgid/sympy/7968e52a-98d9-411b-987c-b26ab7be0059%40googlegroups.com.
> For more options, visit https://groups.google.com/d/optout.

[Jason Moore]

Automatic Differentiation is not applicable to symbolics, see: https://en.wikipedia.org/wiki/Automatic_differentiation

moorepants.info
+01 530-601-9791

> email to sympy+unsubscribe@googlegroups.com.

> To post to this group, send email to sy...@googlegroups.com.
> Visit this group at https://groups.google.com/group/sympy.
> To view this discussion on the web visit
> https://groups.google.com/d/msgid/sympy/7968e52a-98d9-411b-987c-b26ab7be0059%40googlegroups.com.
> For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "sympy" group.

To unsubscribe from this group and stop receiving emails from it, send an email to sympy+unsubscribe@googlegroups.com.

To post to this group, send email to sy...@googlegroups.com.
Visit this group at https://groups.google.com/group/sympy.

To view this discussion on the web visit https://groups.google.com/d/msgid/sympy/CAKgW%3D6LUd4H%3Dsnwkzr9yUn23XPSF2SoirF4DSHd34suwHFF2fg%40mail.gmail.com.