GeezerVison is showing github.com — Return type of eigenvalues from lqr and lqe not consistent · Issue #439 · python-control/python-control
Return type of eigenvalues from lqr and lqe not consistent #439DescriptionThis came up while pytestifying statefbk_test.py (#438)
lqr() will always return E as (nstates, ) array
|
E = w[0:nstates]; |
|
|
|
return _ssmatrix(K), _ssmatrix(S), E |
lqe() returns _ssmatrix(E) which is a (nstates, ) array or (1, nstates) matrix depending on config.defaults.use_numpy_matrix()
|
return _ssmatrix(LT.T), _ssmatrix(P), _ssmatrix(E) |
Activitysawyerbfuller commented on Jul 29, 2020 I am not sure what the convention is here but I saw elsewhere that there
were functions that returned a matrix or an array depending on that
setting, using _ssmatrix. So my take is that Lqr should probably do the
same. S
bnavigator commented on Jul 29, 2020 ContributorAuthor More actions bnavigator commented on Aug 16, 2020 ContributorAuthor More actions murrayrm commented on Dec 24, 2020 I'd like to clean this up before 0.8.4 release, but there are a few issues that could break users' code:
lqr returns _ssmatrix(K), _ssmatrix(S), E, for which the first return value is the gain matrix (2D array or matrix, depending on config settings), the second return value is the solution to the Riccati equation (2D array or matrix), and the third value is a 1D array (list of closed loop eigenvalues).lqe returns _ssmatrix(K), _ssmatrix(S), _ssmatrix(E), which forces the third argument to be a 2D array.care and dare return _ssmatrix(X), w[:n], _ssmatrix(G), for which the first return value is the solution to the Riccati equation (2D array or matrix), the second return value is a list of eigenvalues (1D array), and the third argument is the gain matrix (2D array or matrix).
I think we should make the following change in 0.8.4:
- Change return values for
lqe to match lqr
This will require an updated unit test (as noted above) and could break some users' code. But the fix to user code is simple (convert to 2D array) and so I think it is OK for 0.8.4.
I think we might want to consider whether to make changes to care and dare in 0.9.0:
-
We can leave things the same with the argument that care and dare are returning solutions to the Riccati equation as their primary usage => that solution should be first. The lqr and lqe functions call care or dare, but are returning controllers as their primary usage => OK for the order to be different.
-
We can change the order of the return values for care and dare to match lqr and lqe.
My current feeling is that we leave care and dare alone. Thoughts? sawyerbfuller commented on Dec 24, 2020 Agree with richards’ take that ok for care and dare to have different
return order. Off the top of my head I think matlab might do the same.
Re lqe return argument - I think I made a mistake when I originally wrote
it to naively turn E into an ssmatrix. Agree it should not have been. S
On Thu, Dec 24, 2020 at 9:37 AM Richard Murray ***@***.***> wrote:
I'd like to clean this up before 0.8.4 release, but there are a few issues
that could break users' code:
- lqr returns _ssmatrix(K), _ssmatrix(S), E, for which the first
return value is the gain matrix (2D array or matrix, depending on config
settings), the second return value is the solution to the Riccati equation
(2D array or matrix), and the third value is a 1D array (list of closed
loop eigenvalues).
- lqe returns _ssmatrix(K), _ssmatrix(S), _ssmatrix(E), which forces
the third argument to be a 2D array.
- care and dare return _ssmatrix(X), w[:n], _ssmatrix(G), for which
the first return value is the solution to the Riccati equation (2D array or
matrix), the second return value is a list of eigenvalues (1D array), and
the third argument is the gain matrix (2D array or matrix).
I think we should make the following change in 0.8.4:
- Change return values for lqe to match lqr
This will require an updated unit test (as noted above) and could break
some users' code. But the fix to user code is simple (convert to 2D array)
and so I think it is OK for 0.8.4.
I think we might want to consider whether to make changes to care and dare
in 0.9.0:
-
We can leave things the same with the argument that care and dare are
returning solutions to the Riccati equation as their primary usage => that
solution should be first. The lqr and lqe functions call care or dare,
but are returning controllers as their primary usage => OK for the order to
be different.
-
We can change the order of the return values for care and dare to
match lqr and lqe.
My current feeling is that we leave care and dare alone. Thoughts?
—
You are receiving this because you commented.
Reply to this email directly, view it on GitHub
< #439 (comment)>,
or unsubscribe
< https://github.com/notifications/unsubscribe-auth/AN74SSLASGB633QJ46ETPLTSWN3ZFANCNFSM4PLCB64A>
.
--
Sawyer Fuller
Assistant Professor of Mechanical Engineering
Adjunct, Paul G Allen School of Computer Science
University of Washington
http://faculty.washington.edu/~minster/
(Typed with my thumbs) murrayrm commented on Dec 26, 2020
MetadataMetadataDevelopmentNo branches or pull requestsIssue actions
|