Evaluation

[MaxOstrowski]

Some evaluation of different flattening methods.

• flatzingo
• chuffed
• std (no global constraints)
  numconstraints.zip

bool and int represent number of variables produced by flattening.
Other values indicated the number of specific constraints.

[MaxOstrowski]

Data not complete due to parse error:
potassco/fzn2lp#12

[MaxOstrowski]

results.zip

First results.

1. Base computation speed seems to be quite good in comparison with chuffed
2. For a lot of benchmarks we are lacking global constraints and flatzinc transformation produces too much facts to handle.
3. It is hard to find a good configuration for the competition as benchmarks change every year.
   On last years results (attached) --trans-opt seems to be good and maybe a bit more translation.
   I will probably just use --trans-opt in the competition.
   I plan to do the competition entry early next week, so in case anything else comes to your mind, leave me a note.

[tortinator]

Merci @MaxOstrowski ! I checked the tables and was mainly interested in the comparison to clingcon3 But it seems that this is not in there, right?

[MaxOstrowski]

Not in there. Will try to add something... but this could take some time.

[rkaminsk]

Can we have cactus plots? All those numbers just give me a headache. It takes too much time to rearrange the file to see something. If what you say is true, then there should be parts of the graph where clingcon is below chuffed and some where clingcon is above.

[MaxOstrowski]

results.zip

Unfortunately clingcon-3.3.0 seems to be still quite a bit faster.
Now I consider entering with clingcon3 and clingcon5 into the minizinc competition.
Sorry, not cactus plots yet, data is very messy and needs some cleanup before I could do so. Maybe next week.