mcmasteru/hashedexpression

Type-safe modelling DSL, symbolic transformation, and code generation for solving optimization problems.

hashedexpression's Issues

undefined reference to `var_shape` in ipopt template

At least when working with VNum only values var_shape doesn't seem to be generated in problem.c, so when compiling with ipopt.c we get an undefined reference

Change template to reuse computed value in objective / partial derivatives

Right now it appears templates like ipopt.c don't use evaluate_partial_derivates_and_objective, meaning that shared sub-expressions get recomputed inbetween calculating the objective / partials. Solution: pass evaluate_partial_derivatives_and_objective for both and use a global flag to check if it's already been computed per iteration?

Add maximize keyword to symphony?

because otherwise we're not the cleanest language apparently

support subexpression sharing for Ipopt template

We generate a method of doing combined computation, but we don't actually use it to help make Ipopt more efficient.

@dalvescb, to this end, we might be able to take advantage the new_x Boolean value provided by Ipopt to the eval_f, eval_grad_f, eval_g, eval_jac_g and eval_h functions, documented here as "false if any evaluation method (eval_*) was previously called with the same values in x, true otherwise. This can be helpful when users have efficient implementations that calculate multiple outputs at once. Ipopt internally caches results from the TNLP and generally, this flag can be ignored."

All ipopt problems fail with "Invalid number in NLP function or derivative detected"

can recreate easily by using the "entropy.sp" example in the README. Not sure at what point this broke, but was somewhere inbetween when the haddock-codegen branch was created

Clean Up Branches

Too many branches!!!!