zhaoyuchen100
commented
6 years ago Hi Joao Thanks for your reply. I agree with your intention to make the code as modular as possible. Regarding the bug, it is a little abstract to trace it down from my side, but I believe there is nothing wrong with the learn_lambda_ccl code itself. Might be the data prepared is not ok. What's your suggestion to merge the functions? Best, Yuchen
发自我的 iPhone
在 2017年11月11日，下午12:08，joaomoura24 notifications@github.com 写道：
When trying to compare the estimated projection matrix using our latest method (implemented in def_constraint_estimator.m) and the method implemented in model_lambda_ccl.m I am getting a bug concerning the dimensions of matrix multiplication:
Error in learn_lambda_ccl (line 60) Vn(:,n) = J(X(:,n)) * Un(:,n) ;
The idea is that this 2 methods for constraint estimation should interchange because, at least according documentation, both estimate the null space projection given a constraint modelled as a linear combination of regressors, and provided the regressors. No specific conditions on these regressors are advertised.
When trying to fix this, it generates an error regarding dimensions elsewhere and I couldn't track what exactly is causing this. I don't know if this is due to:
wrong set up of the settings structure; bug in the code; or some assumption on the regressors that I did't realise. The file demo_N_estimate_comp.m in the demos folder runs the set up and the 2 methods for the a synthetic data sets of the kuka arm performing circular motions.
To generate this data set we have to run the file generate_data_kuka_wiping.m (this requires the installation of the MatLab Peter Corke Robotic toolbox through http://petercorke.com/wordpress/toolboxes/robotics-toolbox) in the data_generation folder.
I would greatly appreciate some help on trying to figure out the error source.
My intension would be from now on to start interchanging our methods such constraint and policy estimation and data sets to make sure the code becomes as modular as possible and also to make us understand what are the components of code that are basically doing the same thing (and should be merged) and the components that are different - and how can we compare them. Kind regards,
Joao Moura
— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub, or mute the thread.
joaomoura24
When trying to compare the estimated projection matrix using our latest method (implemented in def_constraint_estimator.m) and the method implemented in model_lambda_ccl.m I am getting a bug concerning the dimensions of matrix multiplication:
Error in learn_lambda_ccl (line 60) Vn(:,n) = J(X(:,n)) * Un(:,n) ;
The idea is that this 2 methods for constraint estimation should interchange because, at least according documentation, both estimate the null space projection given a constraint modelled as a linear combination of regressors, and provided the regressors. No specific conditions on these regressors are advertised.
When trying to fix this, it generates an error regarding dimensions elsewhere and I couldn't track what exactly is causing this. I don't know if this is due to:
The file demo_N_estimate_comp.m in the demos folder runs the set up and the 2 methods for the a synthetic data sets of the kuka arm performing circular motions.
To generate this data set we have to run the file generate_data_kuka_wiping.m (this requires the installation of the MatLab Peter Corke Robotic toolbox through http://petercorke.com/wordpress/toolboxes/robotics-toolbox) in the data_generation folder.
I would greatly appreciate some help on trying to figure out the error source.
My intension would be from now on to start interchanging our methods such constraint and policy estimation and data sets to make sure the code becomes as modular as possible and also to make us understand what are the components of code that are basically doing the same thing (and should be merged) and the components that are different - and how can we compare them. Kind regards,
Joao Moura