Here a list to what I'm doing:

Irace Model tests: Test of changes in the model for categorical parameters. A multiplier is applied after at the end of each iteration based on the survivors of each iteration, the survivors can be defined as the ones alive at the end of the iteration or the ones that survive a test performed during the iteration.
The experiments consist in 20 irace runs performed for each test. The candidates obtained are compared later using a testing set. Comparing graphs and statistical tests can be found here [acotsp, spear, moaco]
Results were not very conclusive, for acotsp show an improvement, but for the other benchmarks is not clear.
  • Moaco elite cadidates: A testing over the training set using all the elite candidates (obtained in the 20 runs described previously) was performed, in order to find out the reason why the models were not improving the model. Testing results and graphs



Tuners Comparison:Different benchmarks using available tuners under different configurations. Comparing graphs can be found here [acotsp-ftest, acotsp-ttest, spear-ftest,spear-ttest, moaco].



SCIP Tuning: Tuning of scip solving the toll problem (Alessia) using generated instances [training / testing]. Comparing graphs can be found here.



SMAC + CPLEX : Tuning of smac using cplex testing with the amount of instances.
  • Instances CAT 100: No apreciable difference in the tuning with all instances or only one.
  • Instances CAT 200: Testing stopped to perform smac other tests in node 4.



Irace parameter Analisys : >Analisys some irace parameters (To be started).


Irace numerical parameters model : Modification of the irace model for numerical parameters! (reading CMAES).