BFGS inverse Hessian

49 views
Skip to first unread message

pa...@soe.ucsc.edu

unread,
May 23, 2017, 4:54:37 PM5/23/17
to Ceres Solver


Hi Everyone,

It is useful to have access to the inverse Hessian matrix while using the BFGS line search direction type unconstrained minimization.

Is there a way to access this member of the BFGS class from the API?

Best,
--Panos

Sameer Agarwal

unread,
May 23, 2017, 4:56:11 PM5/23/17
to Ceres Solver
Panos,
Do you mean the final estimate after the call to Solve?
If so, Alex has a prototype


Sameer


--
You received this message because you are subscribed to the Google Groups "Ceres Solver" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ceres-solver...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/ceres-solver/32317573-e7c0-4464-a9bf-6d762f6a9c13%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

Panos Lambrianides

unread,
May 23, 2017, 5:01:35 PM5/23/17
to ceres-...@googlegroups.com
Yes that is correct-I need the final estimate.  Sorry for not being clear.  Thank you for the code snippet.  

I also hacked the code to do the same but it looks like Alex's solution is more elegant.
Just as a comment this is a useful feature for research purposes.

Best,
--P.

On Tue, May 23, 2017 at 1:55 PM, 'Sameer Agarwal' via Ceres Solver <ceres-...@googlegroups.com> wrote:
Panos,
Do you mean the final estimate after the call to Solve?
If so, Alex has a prototype


Sameer


On Tue, May 23, 2017 at 1:54 PM <pa...@soe.ucsc.edu> wrote:


Hi Everyone,

It is useful to have access to the inverse Hessian matrix while using the BFGS line search direction type unconstrained minimization.

Is there a way to access this member of the BFGS class from the API?

Best,
--Panos

--
You received this message because you are subscribed to the Google Groups "Ceres Solver" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ceres-solver+unsubscribe@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/ceres-solver/32317573-e7c0-4464-a9bf-6d762f6a9c13%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to a topic in the Google Groups "Ceres Solver" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/ceres-solver/RtIEl-qL7ys/unsubscribe.
To unsubscribe from this group and all its topics, send an email to ceres-solver+unsubscribe@googlegroups.com.
 To view this discussion on the web visit https://groups.google.com/d/msgid/ceres-solver/CABqdRUAUXfnHb6%2BcPVvcfrr6JxgaS14tT0NQ%2B%3DLOdANWi2GZ7Q%40mail.gmail.com.

For more options, visit https://groups.google.com/d/optout.



--

Sameer Agarwal

unread,
May 23, 2017, 5:02:29 PM5/23/17
to ceres-...@googlegroups.com
Panos, 
What are you using this for?
Sameer


On Tue, May 23, 2017 at 2:01 PM Panos Lambrianides <pa...@soe.ucsc.edu> wrote:
Yes that is correct-I need the final estimate.  Sorry for not being clear.  Thank you for the code snippet.  

I also hacked the code to do the same but it looks like Alex's solution is more elegant.
Just as a comment this is a useful feature for research purposes.

Best,
--P.

On Tue, May 23, 2017 at 1:55 PM, 'Sameer Agarwal' via Ceres Solver <ceres-...@googlegroups.com> wrote:
Panos,
Do you mean the final estimate after the call to Solve?
If so, Alex has a prototype


Sameer


On Tue, May 23, 2017 at 1:54 PM <pa...@soe.ucsc.edu> wrote:


Hi Everyone,

It is useful to have access to the inverse Hessian matrix while using the BFGS line search direction type unconstrained minimization.

Is there a way to access this member of the BFGS class from the API?

Best,
--Panos

--
You received this message because you are subscribed to the Google Groups "Ceres Solver" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ceres-solver...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/ceres-solver/32317573-e7c0-4464-a9bf-6d762f6a9c13%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to a topic in the Google Groups "Ceres Solver" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/ceres-solver/RtIEl-qL7ys/unsubscribe.
To unsubscribe from this group and all its topics, send an email to ceres-solver...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/ceres-solver/CABqdRUAUXfnHb6%2BcPVvcfrr6JxgaS14tT0NQ%2B%3DLOdANWi2GZ7Q%40mail.gmail.com.

For more options, visit https://groups.google.com/d/optout.



--

--
You received this message because you are subscribed to the Google Groups "Ceres Solver" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ceres-solver...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/ceres-solver/CABgE8ghgOrdOofef7z%2B8kqGnZYSimJo9grhpbTHqzO37Y8_1zA%40mail.gmail.com.

Panos Lambrianides

unread,
May 23, 2017, 5:18:49 PM5/23/17
to ceres-...@googlegroups.com
I am approximating a high dimensional function with a gaussian and then finding the minimum of the negative log likelihood.  I would like to get the covariance matrix without explicitly inverting the Hessian, if possible, because that is an expensive operation.

Sameer Agarwal

unread,
May 23, 2017, 5:22:31 PM5/23/17
to ceres-...@googlegroups.com
and it is okay that the approximation to the covariance matrix is so low rank?

On Tue, May 23, 2017 at 2:18 PM Panos Lambrianides <pa...@soe.ucsc.edu> wrote:
I am approximating a high dimensional function with a gaussian and then finding the minimum of the negative log likelihood.  I would like to get the covariance matrix without explicitly inverting the Hessian, if possible, because that is an expensive operation.

--
You received this message because you are subscribed to the Google Groups "Ceres Solver" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ceres-solver...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/ceres-solver/CABgE8ggesizLd2r5d5nWir7qJj7xRDCEPK%2B17M9Zs027w%2BiQTA%40mail.gmail.com.

Panos Lambrianides

unread,
May 23, 2017, 5:28:57 PM5/23/17
to ceres-...@googlegroups.com
It may be OK for my purposes yes.  I am not sure yet which is why I am trying it out.

On Tue, May 23, 2017 at 2:22 PM, 'Sameer Agarwal' via Ceres Solver <ceres-...@googlegroups.com> wrote:
and it is okay that the approximation to the covariance matrix is so low rank?

On Tue, May 23, 2017 at 2:18 PM Panos Lambrianides <pa...@soe.ucsc.edu> wrote:
I am approximating a high dimensional function with a gaussian and then finding the minimum of the negative log likelihood.  I would like to get the covariance matrix without explicitly inverting the Hessian, if possible, because that is an expensive operation.

 --
You received this message because you are subscribed to the Google Groups "Ceres Solver" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ceres-solver+unsubscribe@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/ceres-solver/CABgE8ggesizLd2r5d5nWir7qJj7xRDCEPK%2B17M9Zs027w%2BiQTA%40mail.gmail.com.
For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to a topic in the Google Groups "Ceres Solver" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/ceres-solver/RtIEl-qL7ys/unsubscribe.
To unsubscribe from this group and all its topics, send an email to ceres-solver+unsubscribe@googlegroups.com.
 To view this discussion on the web visit https://groups.google.com/d/msgid/ceres-solver/CABqdRUAbc%2B9oq02ZinoW4mAVbY4uNeEziEj59d_D%3DWRSZ8oU3A%40mail.gmail.com.

For more options, visit https://groups.google.com/d/optout.

Sameer Agarwal

unread,
May 23, 2017, 5:31:05 PM5/23/17
to ceres-...@googlegroups.com
Please let us know if this turns out to be useful that will help us decide if and how this patch should be added to ceres.
Sameer


On Tue, May 23, 2017 at 2:28 PM Panos Lambrianides <pa...@soe.ucsc.edu> wrote:
It may be OK for my purposes yes.  I am not sure yet which is why I am trying it out.

On Tue, May 23, 2017 at 2:22 PM, 'Sameer Agarwal' via Ceres Solver <ceres-...@googlegroups.com> wrote:
and it is okay that the approximation to the covariance matrix is so low rank?

On Tue, May 23, 2017 at 2:18 PM Panos Lambrianides <pa...@soe.ucsc.edu> wrote:
I am approximating a high dimensional function with a gaussian and then finding the minimum of the negative log likelihood.  I would like to get the covariance matrix without explicitly inverting the Hessian, if possible, because that is an expensive operation.

 --
You received this message because you are subscribed to the Google Groups "Ceres Solver" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ceres-solver...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/ceres-solver/CABgE8ggesizLd2r5d5nWir7qJj7xRDCEPK%2B17M9Zs027w%2BiQTA%40mail.gmail.com.
For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to a topic in the Google Groups "Ceres Solver" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/ceres-solver/RtIEl-qL7ys/unsubscribe.
To unsubscribe from this group and all its topics, send an email to ceres-solver...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/ceres-solver/CABqdRUAbc%2B9oq02ZinoW4mAVbY4uNeEziEj59d_D%3DWRSZ8oU3A%40mail.gmail.com.

For more options, visit https://groups.google.com/d/optout.

--

--
You received this message because you are subscribed to the Google Groups "Ceres Solver" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ceres-solver...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/ceres-solver/CABgE8gjTJ6xnWofc1KB5Lh0prwOVvu__5BwVrQApTUsfR%2B93TA%40mail.gmail.com.

Panos Lambrianides

unread,
May 23, 2017, 5:38:16 PM5/23/17
to ceres-...@googlegroups.com
Sorry for the back and forth.  My approach was inspired by this paper which indicates that it has already been useful for other researchers.  



On Tue, May 23, 2017 at 2:30 PM, 'Sameer Agarwal' via Ceres Solver <ceres-...@googlegroups.com> wrote:
Please let us know if this turns out to be useful that will help us decide if and how this patch should be added to ceres.
Sameer


On Tue, May 23, 2017 at 2:28 PM Panos Lambrianides <pa...@soe.ucsc.edu> wrote:
It may be OK for my purposes yes.  I am not sure yet which is why I am trying it out.

On Tue, May 23, 2017 at 2:22 PM, 'Sameer Agarwal' via Ceres Solver <ceres-...@googlegroups.com> wrote:
and it is okay that the approximation to the covariance matrix is so low rank?

On Tue, May 23, 2017 at 2:18 PM Panos Lambrianides <pa...@soe.ucsc.edu> wrote:
I am approximating a high dimensional function with a gaussian and then finding the minimum of the negative log likelihood.  I would like to get the covariance matrix without explicitly inverting the Hessian, if possible, because that is an expensive operation.

 --
You received this message because you are subscribed to the Google Groups "Ceres Solver" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ceres-solver+unsubscribe@googlegroups.com.

--
You received this message because you are subscribed to a topic in the Google Groups "Ceres Solver" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/ceres-solver/RtIEl-qL7ys/unsubscribe.
To unsubscribe from this group and all its topics, send an email to ceres-solver+unsubscribe@googlegroups.com.

For more options, visit https://groups.google.com/d/optout.

--

--
You received this message because you are subscribed to the Google Groups "Ceres Solver" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ceres-solver+unsubscribe@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/ceres-solver/CABgE8gjTJ6xnWofc1KB5Lh0prwOVvu__5BwVrQApTUsfR%2B93TA%40mail.gmail.com.
For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to a topic in the Google Groups "Ceres Solver" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/ceres-solver/RtIEl-qL7ys/unsubscribe.
To unsubscribe from this group and all its topics, send an email to ceres-solver+unsubscribe@googlegroups.com.
 To view this discussion on the web visit https://groups.google.com/d/msgid/ceres-solver/CABqdRUBoyxHDguuv%2B-rGeZFWDWnxXM1S8hipQUAop%3Do-%3DNEQ%2Bg%40mail.gmail.com.

For more options, visit https://groups.google.com/d/optout.

Sameer Agarwal

unread,
May 23, 2017, 7:55:19 PM5/23/17
to ceres-...@googlegroups.com
Thanks for the link
Let us know if it works for you, and alex and I will take it from there.

On Tue, May 23, 2017 at 2:38 PM Panos Lambrianides <pa...@soe.ucsc.edu> wrote:
Sorry for the back and forth.  My approach was inspired by this paper which indicates that it has already been useful for other researchers.  



On Tue, May 23, 2017 at 2:30 PM, 'Sameer Agarwal' via Ceres Solver <ceres-...@googlegroups.com> wrote:
Please let us know if this turns out to be useful that will help us decide if and how this patch should be added to ceres.
Sameer


On Tue, May 23, 2017 at 2:28 PM Panos Lambrianides <pa...@soe.ucsc.edu> wrote:
It may be OK for my purposes yes.  I am not sure yet which is why I am trying it out.

On Tue, May 23, 2017 at 2:22 PM, 'Sameer Agarwal' via Ceres Solver <ceres-...@googlegroups.com> wrote:
and it is okay that the approximation to the covariance matrix is so low rank?

On Tue, May 23, 2017 at 2:18 PM Panos Lambrianides <pa...@soe.ucsc.edu> wrote:
I am approximating a high dimensional function with a gaussian and then finding the minimum of the negative log likelihood.  I would like to get the covariance matrix without explicitly inverting the Hessian, if possible, because that is an expensive operation.

 --
You received this message because you are subscribed to the Google Groups "Ceres Solver" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ceres-solver...@googlegroups.com.

--
You received this message because you are subscribed to a topic in the Google Groups "Ceres Solver" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/ceres-solver/RtIEl-qL7ys/unsubscribe.
To unsubscribe from this group and all its topics, send an email to ceres-solver...@googlegroups.com.

For more options, visit https://groups.google.com/d/optout.

--

--
You received this message because you are subscribed to the Google Groups "Ceres Solver" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ceres-solver...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/ceres-solver/CABgE8gjTJ6xnWofc1KB5Lh0prwOVvu__5BwVrQApTUsfR%2B93TA%40mail.gmail.com.
For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to a topic in the Google Groups "Ceres Solver" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/ceres-solver/RtIEl-qL7ys/unsubscribe.
To unsubscribe from this group and all its topics, send an email to ceres-solver...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/ceres-solver/CABqdRUBoyxHDguuv%2B-rGeZFWDWnxXM1S8hipQUAop%3Do-%3DNEQ%2Bg%40mail.gmail.com.

For more options, visit https://groups.google.com/d/optout.



--

--
You received this message because you are subscribed to the Google Groups "Ceres Solver" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ceres-solver...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/ceres-solver/CABgE8gi8AeHHBNsuR10zHUvaR5%2BD2mZpKLZQum7uWtwQc3wR%3Dg%40mail.gmail.com.
Reply all
Reply to author
Forward
0 new messages