Implementation of first version #1
Conversation
Gosh this was painful.
Now it inherits from Evaluable.
This allows to output an expression as derivative.
…orsEvaluables. Missing row and element evaluables.
Commented the derivative parts to test them one by one.
It checks whether an evaluable is dependent on a variable.
|
Cool. What's the plan? It will be a dependency of iDynTree? Or of the dynamic planner code? |
|
Do you want to move the repo to |
|
The plan would be to test it first inside the planner, also to check whether it is already enough for the use cases I need. Afterwards I was thinking to move it in one of our organizations. |
In view of future move to |
|
I would set up the CI with at least Visual Studio as soon as possible, such template heavy code is typically a good way to encover bugs in compilers. : ) |
Definitely agree. Do you have any entry point for this? |
Added some debug prints.
Allowing to specify the name of generics, helpers and commons.
… not when compiling)
…bit the time to compute the squeeze evaluable. Improved definitionof the operator == and !=.
This remove the need of setting to false all the entries of the register in case of new values.
The concept of Registrar is introduced, whose aim is to avoid having to go through all the expression tree to know if the expression is new. Indeed, only few evaluables can change their values. In particular, these are variables and mutables. The latter are similar to constants with the difference that they have defined the operator equal. All the methods isNew and isDependent from have been substituted by the method addDependencies where the actual dependencies are specified.
…e computational cost.
Having raw pointers allows to store (this) as dependency. In turn this allowed to define the Assignable object, where the operator = is defined, while avoiding to override the getDependencies method. In turn, the static check for the = operator is easier. This seemed to be causing issues in VCC and clang.
… the dependencies.
|
This PR has definitively gone out of control. Sorry for the spam. Merging as is, facing the remaining problems in other issues. |
This PR implements the first version of the
sDifftool. This tool mimics automatic differentiation for expressions where known blocks are involved.The code is built around Eigen matrices, uses templates extensively and is header only.
A proper readme is missing.
I am open to suggestions for a fancier name.
FYI
@DanielePucci, @traversaro, @GiulioRomualdi, @diegoferigo
TODO