Select problems

[abelsiqueira]

This implements a classification of the CUTEst problems using $MASTSIF/CLASSF.DB and the problem itself opened in Julia.

This also implements a select tool based on that classification. It should be useful along with JuliaSmoothOptimizers/Optimize.jl

This also uncovered some problems with the MASTSIF classification, which can be seen at the travis logs.

[coveralls]

Coverage increased (+0.2%) to 64.052% when pulling 05c98e8 on abelsiqueira:feat/class into 58dc340 on JuliaSmoothOptimizers:develop.

[dpo]

sifdecoder should probably also take an AbstractString as input?

[abelsiqueira]

The alternative to ASCIIString is UTFxString, and since I don't know how the command line sifdecoder will handle that, I made that change. If you think the alternative is better, I'll make the change

[dpo]

What I'm saying is that the Julia function sifdecoder() should probably accept an AbstractString. I presume string interpolation does the right thing.

[dpo]

This is a lot of work! How do you envision problem selection should go for other problem sources (AMPL, JuMP, Simple)?

[abelsiqueira]

I think the only other selection we need to worry is the JuliaSmoothOptimizers/OptimizationProblems.jl#3. CUTEst is a special case among the NLPModels, because it has the repository as main objective.

This branch shows a possible course of action, for OptimizationProblems. We define static characteristics that are not easy to verify (like the MASTSIF table), and query other characteristics creating a .json file (or a database?) and update the json every time there is a new problem. The selection function opens the json file, which is relatively fast.
Advantages (change json to another DB, if desired)

• Once the json is created, querying is relatively fast.
• The json can be used by other softwares/packages (online selector?)
• If we define a json standard, a simgle NLPQuery can be created.

Disavantages

• Every new problem needs a manual update.
• Demands additional static characteristics.

[dpo]

Right. I like the JSON table, the flexibility it offers and how it's a lot more future-proof than the CLASSF.DB file. I'm thinking a SIF-style classification string could be supplemented with data we gather by instantiating problems. The meta data structure contains a lot of info. For example, CLASSF.DB currently doesn't let you select equality-constrained problems only or inequality-constrained problems only.

[abelsiqueira]

Updated

[coveralls]

Coverage increased (+0.2%) to 64.071% when pulling bf34dd5 on abelsiqueira:feat/class into 58dc340 on JuliaSmoothOptimizers:develop.

[coveralls]

Coverage increased (+0.4%) to 53.705% when pulling dc1d1c9 on abelsiqueira:feat/class into 4c72dac on JuliaSmoothOptimizers:develop.

[coveralls]

Coverage increased (+0.4%) to 53.705% when pulling 423bc2d on abelsiqueira:feat/class into 4c72dac on JuliaSmoothOptimizers:develop.

[coveralls]

Coverage increased (+0.4%) to 53.705% when pulling a96a7e2 on abelsiqueira:feat/class into 4c72dac on JuliaSmoothOptimizers:develop.

[dpo]

I'm quite a bit behind reviewing this. @vepiteski It would be great if you could also test this.

[dpo]

NB: There's a new ready to go label that you can apply to PRs to indicate that you're finished with the changes.

[coveralls]

Coverage increased (+0.4%) to 53.705% when pulling 479ffd7 on abelsiqueira:feat/class into 34e368f on JuliaSmoothOptimizers:develop.

[coveralls]

Coverage increased (+0.4%) to 53.705% when pulling 479ffd7 on abelsiqueira:feat/class into 34e368f on JuliaSmoothOptimizers:develop.

[coveralls]

Coverage increased (+0.4%) to 53.705% when pulling 479ffd7 on abelsiqueira:feat/class into 34e368f on JuliaSmoothOptimizers:develop.

[coveralls]

Coverage increased (+0.4%) to 53.705% when pulling 479ffd7 on abelsiqueira:feat/class into 34e368f on JuliaSmoothOptimizers:develop.

[coveralls]

Coverage increased (+0.4%) to 53.705% when pulling 22b6388 on abelsiqueira:feat/class into 34e368f on JuliaSmoothOptimizers:develop.

[abelsiqueira]

Rebased

[coveralls]

Coverage increased (+0.6%) to 58.119% when pulling 4937c2d on abelsiqueira:feat/class into 7775008 on JuliaSmoothOptimizers:develop.

[coveralls]

Coverage increased (+0.6%) to 58.119% when pulling 4937c2d on abelsiqueira:feat/class into 7775008 on JuliaSmoothOptimizers:develop.

[timholy]

FYI this is working very well for me.

[dpo]

Thanks. I'm swamped and can't find a chance to test. I propose we merge this so others can start using it. Many thanks @abelsiqueira !