[jira] [Commented] (MATH-1394) Implementation of DIRECT global optimizer
[ https://issues.apache.org/jira/browse/MATH-1394?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17739426#comment-17739426 ] Gilles Sadowski commented on MATH-1394: --- bq. I was wondering if I can use the clustering apis that common-maths have Sure; it's always better if we can reuse existing code. Note that there are some issues with the design of the clustering functionality (reported on JIRA). Eventually, fixing those would allow the refactored code to be moved to a new (maven) module. > Implementation of DIRECT global optimizer > - > > Key: MATH-1394 > URL: https://issues.apache.org/jira/browse/MATH-1394 > Project: Commons Math > Issue Type: New Feature >Affects Versions: 3.5, 3.6 >Reporter: Kevin A. Burton >Priority: Major > Fix For: 4.X > > Attachments: DIRECT.png, GLOBAL.png > > > An open source implementation of the DIRECT global optimizer, described by > Jones, Perttunen and Stuckmann, implementing > math3.optim.nonlinear.scalar.MultivariateOptimizer, is available as > DIRECTOptimizer.java at > https://github.com/edwardkort/WWIDesigner/tree/optimizer/WWIDesigner/src/main/com/wwidesigner/math. > There are also three variants of the algorithm on that page: > DIRECT_L_Optimizer implements DIRECT-L by Gablonsky and Kelley, > DIRECT1Optimizer changes which sides are chosen for dividing, and > DIRECTCOptimizer adds alternative ways to select potentially-optimal > hyperrectangles. > DIRECT is not as fast as BOBYQA, but is better at finding a global minimum in > a field of many local minima. > JUnit tests for all of these, using standard optimizer test functions, are > available in > https://github.com/edwardkort/WWIDesigner/tree/optimizer/WWIDesigner/src/test/com/wwidesigner/math. -- This message was sent by Atlassian Jira (v8.20.10#820010)
[jira] [Commented] (MATH-1394) Implementation of DIRECT global optimizer
[ https://issues.apache.org/jira/browse/MATH-1394?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17739374#comment-17739374 ] Kiriakos Marantidis commented on MATH-1394: --- So there is a step where we cluster some points. I was wondering if I can use the clustering apis that common-maths have or if I need to do some from scratch implementation. Also since we have multiple times generation of points, sorting of these points and merging of these points with an existing set of points I think the only efficient data structure to handle such data is a Tree. > Implementation of DIRECT global optimizer > - > > Key: MATH-1394 > URL: https://issues.apache.org/jira/browse/MATH-1394 > Project: Commons Math > Issue Type: New Feature >Affects Versions: 3.5, 3.6 >Reporter: Kevin A. Burton >Priority: Major > Fix For: 4.X > > Attachments: DIRECT.png, GLOBAL.png > > > An open source implementation of the DIRECT global optimizer, described by > Jones, Perttunen and Stuckmann, implementing > math3.optim.nonlinear.scalar.MultivariateOptimizer, is available as > DIRECTOptimizer.java at > https://github.com/edwardkort/WWIDesigner/tree/optimizer/WWIDesigner/src/main/com/wwidesigner/math. > There are also three variants of the algorithm on that page: > DIRECT_L_Optimizer implements DIRECT-L by Gablonsky and Kelley, > DIRECT1Optimizer changes which sides are chosen for dividing, and > DIRECTCOptimizer adds alternative ways to select potentially-optimal > hyperrectangles. > DIRECT is not as fast as BOBYQA, but is better at finding a global minimum in > a field of many local minima. > JUnit tests for all of these, using standard optimizer test functions, are > available in > https://github.com/edwardkort/WWIDesigner/tree/optimizer/WWIDesigner/src/test/com/wwidesigner/math. -- This message was sent by Atlassian Jira (v8.20.10#820010)
[jira] [Commented] (MATH-1394) Implementation of DIRECT global optimizer
[
https://issues.apache.org/jira/browse/MATH-1394?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17737141#comment-17737141
]
Gilles Sadowski commented on MATH-1394:
---
You can certainly give it a try.
Eventually, it would be quite useful to compare it with the already implemented
algorithms
([{{PowellOptimizer}}|https://github.com/apache/commons-math/blob/master/commons-math-legacy/src/main/java/org/apache/commons/math4/legacy/optim/nonlinear/scalar/noderiv/PowellOptimizer.java],
[{{CMAESOptimizer}}|https://github.com/apache/commons-math/blob/master/commons-math-legacy/src/main/java/org/apache/commons/math4/legacy/optim/nonlinear/scalar/noderiv/CMAESOptimizer.java],
[{{SimplexOptimizer}}|https://github.com/apache/commons-math/blob/master/commons-math-legacy/src/main/java/org/apache/commons/math4/legacy/optim/nonlinear/scalar/noderiv/SimplexOptimizer.java]).
> Implementation of DIRECT global optimizer
> -
>
> Key: MATH-1394
> URL: https://issues.apache.org/jira/browse/MATH-1394
> Project: Commons Math
> Issue Type: New Feature
>Affects Versions: 3.5, 3.6
>Reporter: Kevin A. Burton
>Priority: Major
> Fix For: 4.X
>
> Attachments: DIRECT.png, GLOBAL.png
>
>
> An open source implementation of the DIRECT global optimizer, described by
> Jones, Perttunen and Stuckmann, implementing
> math3.optim.nonlinear.scalar.MultivariateOptimizer, is available as
> DIRECTOptimizer.java at
> https://github.com/edwardkort/WWIDesigner/tree/optimizer/WWIDesigner/src/main/com/wwidesigner/math.
> There are also three variants of the algorithm on that page:
> DIRECT_L_Optimizer implements DIRECT-L by Gablonsky and Kelley,
> DIRECT1Optimizer changes which sides are chosen for dividing, and
> DIRECTCOptimizer adds alternative ways to select potentially-optimal
> hyperrectangles.
> DIRECT is not as fast as BOBYQA, but is better at finding a global minimum in
> a field of many local minima.
> JUnit tests for all of these, using standard optimizer test functions, are
> available in
> https://github.com/edwardkort/WWIDesigner/tree/optimizer/WWIDesigner/src/test/com/wwidesigner/math.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
[jira] [Commented] (MATH-1394) Implementation of DIRECT global optimizer
[ https://issues.apache.org/jira/browse/MATH-1394?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17737058#comment-17737058 ] Kiriakos Marantidis commented on MATH-1394: --- [~erans] , This is the algorithm that the paper proposes for the Global Algoritm !GLOBAL.png! And this is the Direct implementation for local search !DIRECT.png! If you believe that's an acceptable pseudo-code algorithm for the Direct optimizer I can start working on its implementation. > Implementation of DIRECT global optimizer > - > > Key: MATH-1394 > URL: https://issues.apache.org/jira/browse/MATH-1394 > Project: Commons Math > Issue Type: New Feature >Affects Versions: 3.5, 3.6 >Reporter: Kevin A. Burton >Priority: Major > Fix For: 4.X > > Attachments: DIRECT.png, GLOBAL.png > > > An open source implementation of the DIRECT global optimizer, described by > Jones, Perttunen and Stuckmann, implementing > math3.optim.nonlinear.scalar.MultivariateOptimizer, is available as > DIRECTOptimizer.java at > https://github.com/edwardkort/WWIDesigner/tree/optimizer/WWIDesigner/src/main/com/wwidesigner/math. > There are also three variants of the algorithm on that page: > DIRECT_L_Optimizer implements DIRECT-L by Gablonsky and Kelley, > DIRECT1Optimizer changes which sides are chosen for dividing, and > DIRECTCOptimizer adds alternative ways to select potentially-optimal > hyperrectangles. > DIRECT is not as fast as BOBYQA, but is better at finding a global minimum in > a field of many local minima. > JUnit tests for all of these, using standard optimizer test functions, are > available in > https://github.com/edwardkort/WWIDesigner/tree/optimizer/WWIDesigner/src/test/com/wwidesigner/math. -- This message was sent by Atlassian Jira (v8.20.10#820010)
[jira] [Commented] (MATH-1394) Implementation of DIRECT global optimizer
[ https://issues.apache.org/jira/browse/MATH-1394?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17736759#comment-17736759 ] Kiriakos Marantidis commented on MATH-1394: --- Yes I was referring to the algorithm specified in the paper you posted. I also found this paper [Global Algorithm|http://publicatio.bibl.u-szeged.hu/17746/7/30418895.pdf] which has a lot of peuso-code for each algorithm it analyzes. As I understand the local search defines if the global algorithm is direct and it provides the algorithm for it (Algorithm 2.1 The basic UNIRANDI local search method, page 17 of the pdf). > Implementation of DIRECT global optimizer > - > > Key: MATH-1394 > URL: https://issues.apache.org/jira/browse/MATH-1394 > Project: Commons Math > Issue Type: New Feature >Affects Versions: 3.5, 3.6 >Reporter: Kevin A. Burton >Priority: Major > Fix For: 4.X > > > An open source implementation of the DIRECT global optimizer, described by > Jones, Perttunen and Stuckmann, implementing > math3.optim.nonlinear.scalar.MultivariateOptimizer, is available as > DIRECTOptimizer.java at > https://github.com/edwardkort/WWIDesigner/tree/optimizer/WWIDesigner/src/main/com/wwidesigner/math. > There are also three variants of the algorithm on that page: > DIRECT_L_Optimizer implements DIRECT-L by Gablonsky and Kelley, > DIRECT1Optimizer changes which sides are chosen for dividing, and > DIRECTCOptimizer adds alternative ways to select potentially-optimal > hyperrectangles. > DIRECT is not as fast as BOBYQA, but is better at finding a global minimum in > a field of many local minima. > JUnit tests for all of these, using standard optimizer test functions, are > available in > https://github.com/edwardkort/WWIDesigner/tree/optimizer/WWIDesigner/src/test/com/wwidesigner/math. -- This message was sent by Atlassian Jira (v8.20.10#820010)
[jira] [Commented] (MATH-1394) Implementation of DIRECT global optimizer
[
https://issues.apache.org/jira/browse/MATH-1394?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17736754#comment-17736754
]
Gilles Sadowski commented on MATH-1394:
---
{quote}I see in description there is a link to a paper [...]
{quote}
I don't see the link to such a paper, only a link to GH...
A quick search on Web turned up the link which I
[posted|https://link.springer.com/article/10.1007/s10898-020-00952-6] on
MATH-621.
bq. [Is a] paper acceptable [as the] basis to implement the functionality from
scratch?
Sure.
> Implementation of DIRECT global optimizer
> -
>
> Key: MATH-1394
> URL: https://issues.apache.org/jira/browse/MATH-1394
> Project: Commons Math
> Issue Type: New Feature
>Affects Versions: 3.5, 3.6
>Reporter: Kevin A. Burton
>Priority: Major
> Fix For: 4.X
>
>
> An open source implementation of the DIRECT global optimizer, described by
> Jones, Perttunen and Stuckmann, implementing
> math3.optim.nonlinear.scalar.MultivariateOptimizer, is available as
> DIRECTOptimizer.java at
> https://github.com/edwardkort/WWIDesigner/tree/optimizer/WWIDesigner/src/main/com/wwidesigner/math.
> There are also three variants of the algorithm on that page:
> DIRECT_L_Optimizer implements DIRECT-L by Gablonsky and Kelley,
> DIRECT1Optimizer changes which sides are chosen for dividing, and
> DIRECTCOptimizer adds alternative ways to select potentially-optimal
> hyperrectangles.
> DIRECT is not as fast as BOBYQA, but is better at finding a global minimum in
> a field of many local minima.
> JUnit tests for all of these, using standard optimizer test functions, are
> available in
> https://github.com/edwardkort/WWIDesigner/tree/optimizer/WWIDesigner/src/test/com/wwidesigner/math.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
[jira] [Commented] (MATH-1394) Implementation of DIRECT global optimizer
[ https://issues.apache.org/jira/browse/MATH-1394?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17736750#comment-17736750 ] Kiriakos Marantidis commented on MATH-1394: --- Ok I think I will try finding the description of the algorithm to implement from scratch. Should I put any content I find like a paper here, to further discuss? I see in description there is a link to a paper which analyzes two forms of the algorithm( the original and the re-visioned one). Isn't this paper acceptable to take as a basis to implement the functionality from scratch? > Implementation of DIRECT global optimizer > - > > Key: MATH-1394 > URL: https://issues.apache.org/jira/browse/MATH-1394 > Project: Commons Math > Issue Type: New Feature >Affects Versions: 3.5, 3.6 >Reporter: Kevin A. Burton >Priority: Major > Fix For: 4.X > > > An open source implementation of the DIRECT global optimizer, described by > Jones, Perttunen and Stuckmann, implementing > math3.optim.nonlinear.scalar.MultivariateOptimizer, is available as > DIRECTOptimizer.java at > https://github.com/edwardkort/WWIDesigner/tree/optimizer/WWIDesigner/src/main/com/wwidesigner/math. > There are also three variants of the algorithm on that page: > DIRECT_L_Optimizer implements DIRECT-L by Gablonsky and Kelley, > DIRECT1Optimizer changes which sides are chosen for dividing, and > DIRECTCOptimizer adds alternative ways to select potentially-optimal > hyperrectangles. > DIRECT is not as fast as BOBYQA, but is better at finding a global minimum in > a field of many local minima. > JUnit tests for all of these, using standard optimizer test functions, are > available in > https://github.com/edwardkort/WWIDesigner/tree/optimizer/WWIDesigner/src/test/com/wwidesigner/math. -- This message was sent by Atlassian Jira (v8.20.10#820010)
[jira] [Commented] (MATH-1394) Implementation of DIRECT global optimizer
[
https://issues.apache.org/jira/browse/MATH-1394?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17736747#comment-17736747
]
Gilles Sadowski commented on MATH-1394:
---
{quote}
[...] transfer the functionality from the external repository to commons-math
as well as the tests that come with it.
{quote}
That would be _forbidden_, because of the [incompatible
license|https://www.apache.org/legal/resolved.html#category-x].
We must find either
* another implementation (with a compatible license), or
* a description of the algorithm to be implemented "from scratch".
Of course, a developer is allowed to _run_ the GPL code in order to compare its
output with the one from the new implementation being developed.
> Implementation of DIRECT global optimizer
> -
>
> Key: MATH-1394
> URL: https://issues.apache.org/jira/browse/MATH-1394
> Project: Commons Math
> Issue Type: New Feature
>Affects Versions: 3.5, 3.6
>Reporter: Kevin A. Burton
>Priority: Major
> Fix For: 4.X
>
>
> An open source implementation of the DIRECT global optimizer, described by
> Jones, Perttunen and Stuckmann, implementing
> math3.optim.nonlinear.scalar.MultivariateOptimizer, is available as
> DIRECTOptimizer.java at
> https://github.com/edwardkort/WWIDesigner/tree/optimizer/WWIDesigner/src/main/com/wwidesigner/math.
> There are also three variants of the algorithm on that page:
> DIRECT_L_Optimizer implements DIRECT-L by Gablonsky and Kelley,
> DIRECT1Optimizer changes which sides are chosen for dividing, and
> DIRECTCOptimizer adds alternative ways to select potentially-optimal
> hyperrectangles.
> DIRECT is not as fast as BOBYQA, but is better at finding a global minimum in
> a field of many local minima.
> JUnit tests for all of these, using standard optimizer test functions, are
> available in
> https://github.com/edwardkort/WWIDesigner/tree/optimizer/WWIDesigner/src/test/com/wwidesigner/math.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
[jira] [Commented] (MATH-1394) Implementation of DIRECT global optimizer
[ https://issues.apache.org/jira/browse/MATH-1394?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17736744#comment-17736744 ] Kiriakos Marantidis commented on MATH-1394: --- Hello, [~erans] I would like to contribute on this one but does the issue of the licenses divergence persist? In terms of what needs to be done I understand that we need to transfer the functionality from the external repository to commons-math as well as the tests that come with it. > Implementation of DIRECT global optimizer > - > > Key: MATH-1394 > URL: https://issues.apache.org/jira/browse/MATH-1394 > Project: Commons Math > Issue Type: New Feature >Affects Versions: 3.5, 3.6 >Reporter: Kevin A. Burton >Priority: Major > Fix For: 4.X > > > An open source implementation of the DIRECT global optimizer, described by > Jones, Perttunen and Stuckmann, implementing > math3.optim.nonlinear.scalar.MultivariateOptimizer, is available as > DIRECTOptimizer.java at > https://github.com/edwardkort/WWIDesigner/tree/optimizer/WWIDesigner/src/main/com/wwidesigner/math. > There are also three variants of the algorithm on that page: > DIRECT_L_Optimizer implements DIRECT-L by Gablonsky and Kelley, > DIRECT1Optimizer changes which sides are chosen for dividing, and > DIRECTCOptimizer adds alternative ways to select potentially-optimal > hyperrectangles. > DIRECT is not as fast as BOBYQA, but is better at finding a global minimum in > a field of many local minima. > JUnit tests for all of these, using standard optimizer test functions, are > available in > https://github.com/edwardkort/WWIDesigner/tree/optimizer/WWIDesigner/src/test/com/wwidesigner/math. -- This message was sent by Atlassian Jira (v8.20.10#820010)
[jira] [Commented] (MATH-1394) Implementation of DIRECT global optimizer
[ https://issues.apache.org/jira/browse/MATH-1394?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16859971#comment-16859971 ] Gilles commented on MATH-1394: -- Sorry I somehow missed this report. As it is, the code cannot be imported into the "Commons Math" code base (because of divergence on what constitutes _free_ in the views of Apache vs GNU). I see that you are one of the authors of the referred sources. That functionality will be welcome in "Commons Math". Are you willing to port it (rebased on the current "master" branch) and provide it under ASL v2? > Implementation of DIRECT global optimizer > - > > Key: MATH-1394 > URL: https://issues.apache.org/jira/browse/MATH-1394 > Project: Commons Math > Issue Type: New Feature >Affects Versions: 3.5, 3.6 >Reporter: Burton Patkau >Priority: Major > Fix For: 4.X > > > An open source implementation of the DIRECT global optimizer, described by > Jones, Perttunen and Stuckmann, implementing > math3.optim.nonlinear.scalar.MultivariateOptimizer, is available as > DIRECTOptimizer.java at > https://github.com/edwardkort/WWIDesigner/tree/optimizer/WWIDesigner/src/main/com/wwidesigner/math. > There are also three variants of the algorithm on that page: > DIRECT_L_Optimizer implements DIRECT-L by Gablonsky and Kelley, > DIRECT1Optimizer changes which sides are chosen for dividing, and > DIRECTCOptimizer adds alternative ways to select potentially-optimal > hyperrectangles. > DIRECT is not as fast as BOBYQA, but is better at finding a global minimum in > a field of many local minima. > JUnit tests for all of these, using standard optimizer test functions, are > available in > https://github.com/edwardkort/WWIDesigner/tree/optimizer/WWIDesigner/src/test/com/wwidesigner/math. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
