marcoaaguiar / yaocptool Goto Github PK

Ref Issue #30

Reference #31

Hello~ I'm very interested in this project. However, I encountered some problems when trying this project.
For example, when I run pce_example.py, I meet the following error:

pce_problem = pce_converter.convert_socp_to_ocp_with_pce()
File "D:\programme\Anaconda\envs\multi-agent\lib\site-packages\yaocptool\stochastic\pce.py", line 150, in convert_socp_to_ocp_with_pce
self.model, cost_list = self._create_model(self.sampled_parameters)
File "D:\programme\Anaconda\envs\multi-agent\lib\site-packages\yaocptool\stochastic\pce.py", line 376, in _create_model
model.include_control(self.socp.model.u_sym)
AttributeError: 'SystemModel' object has no attribute 'u_sym'

And other examples maybe have some errors. I'm not sure if the problem is with the installed version, the Dev and Master versions conflict again. Could you help me?

I know it is indicated to use pip to install this package, but today is also very common to use conda as package manager to python, and also so setup environments.

When trying to use conda to install the dependencies for yaocptool I got some conflicting dependencies between them. More specifically,

• casadi is available through the "conda-forge" channel, but is only available for python2.7, 3.5 or 3.6
• sobol_seq is available only through "https://conda.binstar.org/naught101" channel, but requires python3.4

Maybe I am missing something, but anyway could you please provide a manner to install this package through conda?

The nlp problem so far is split in "nlp_probl" dict, "nlp_call" dict, and the solver.
It would be better to have an OptimizationProblem class, possible sub-classes: NonlinearProblem and QuadraticProblem

use "import logging"

Should not the name of the project be Y(et)A(nother)O(ptimal)C(ontrol)TOOL (yaoctool) instead of yaocptool?

use 'x' instead of 'x_sym', the same for the other variables. for compatibility do this using @Property
and @x.setter

Abstract class for:

• SystemModel
• OCP (?)
• SolutionMethod
• MPC
• Estimator
• OptimizationProblem
• Plant

The following methods could be rewritten/eliminated
Methods:

• create_solver
• call_solver
• solve_raw
• solve

Related to #7

• Merge datasets.
• Put gaps on plots where data is missing.
• get_entry (?)
• get_entry_names (?)
• get_entry_size (?)

Currently the u_func and u_par are in the SytemModel class, but there is no method of chaning them, also no way to check if a model is parametrized, and number of control parameters.

Suggestion: create a dictionary with a mapping of variable names to the state vector.

This would avoid a duplicated implementations, for instance of the plot function. Or re-implementation in the case of improvement proposed by Issue #12

This creates an exception "eval_sx" for integrator object (which do not accept SX.sym as input)

Make a conversion function for OptimizationResult to dataset.
The plot function of the dataset could be used by the optimziationresult to avoid reimplementation of the plot function