search

NicMcPhee / XO-bias-study

Results and write-up of our genetic programming crossover bias study
MIT License
0 stars 0 forks source link

Should we increase the max tree depth limit? #12

Open NicMcPhee opened 9 years ago

NicMcPhee commented 9 years ago

The default in koza.params in ECJ is a max tree depth of 17, which the comments there call "a little wimpy". Increasing it will definitely increase memory usage and slow down runs (bigger trees take longer to process) and might improve performance, especially on some of the harder problems.

I'm not entirely convinced about how "wimpy" 17 is, though. For binary trees a depth of 17 gives you tree sizes of approximately 2^17 nodes, which is a pretty hefty set of nodes to work with. I'm inclined to leave things alone for now, but I wanted to raise the issue in case anyone else has thoughts on the matter.

NicMcPhee commented 9 years ago

I think for the moment we'll leave that be. I don't see much/any evidence that 17 is providing any severe limitation, and messing with it would just give us another variable we wouldn't really understand. I'm not going to mark this as closed, but I will move it to a "possible future stuff" milestone instead of in the "Design experiments" milestone.