Make the number of programs for each level for probabilities match the numbers from the research paper

[nicolaefilat]

Our output

Research paper

[sebdumancic]

Just to mention here as well: small differences are ok (and often happen), but the general numbers should be similar if you are implementing the same procedure.

It would be very interesting to see what the numbers (and performance) look like when you use different search techniques.

Did we point you out to an on-going implementation of bottom-up for guided search? https://github.com/Herb-AI/HerbSearch.jl/blob/bottom-up-iterator/src/bottom_up_iterator.jl