Parameters of irace to be tuned

[ndangtt]

(Updated) on Jan 1, 2024

(used with runtime scenarios where maxTime is set, therefore, elitist is also set as 1)

Simple list of parameters to try first

--capping 0/1 --cappingType [median, mean, worst, best] | capping == 1 --boundType [candidate, instance] | capping == 1 --testType f-test/t-test --elitist-limit [1, 10]

Full list (to be considered at a later phase):

(numbered according to their potential importance)

3. firstTest [1,10] (left out for now) eachTest [1,5]
4. testType f-test, t-test (we could add new ones, like sequential halving or different tests)
5. --min-survival [1, 10]
6. --num-configurations 0 or [2, 100] (ordinal, or a dummy parameter) (left out for now) --confidence [55,98] (left out for now) --soft-restart 0 or 1
7. --capping 0 or 1 (left out for now) --elitist-new-instances 0 or 1 | elitist == 1 (left out for now) --elitist-limit 0 or 2 | elitist == 1
8. --capping-type [median, mean, worst, best] | capping == 1
9. --bound-type [candidate, instance] | capping = 1 ?. --elitist-limit [1, 10]

We could add capping options: https://github.com/souzamarcelo/capopt

effort-type: time capping: True penalty: best-so-far envelope: area strategy: adaptive ag: 0.4 epsilon: 0.05

We could add evolutionary operators: https://github.com/NEO-Research-Group/irace-ea

--crossover and --mutation

[MLopez-Ibanez]

If you are using maxTime, remove elitist=0.

You could also add --elitist-limit [1, 10].

[ndangtt]

thank you! We'll add that new one into the list