Hacks and Optimizations

Some strategies that we might want to consider for the final performance figures, in no particular order:

1. Whether to use a naive successor generation strategy or the match-tree strategy.

2. We might want not to pass those nodes having e.g. novelty w_{#g}=1 through the novelty-two tables. This changes the search, of course, since we'll have more nodes with novelty 2 to expand, but might prove beneficial, esp. when only a handful of novelty-2 expansions are needed?

3. When using simulations with propositional domains, we'll want to test whether it's better or not to consider "negative" propositions.

4. Implement a parameter similar to M in H&N ICAPS'17 paper:

In the resulting algorithm, up to M nodes with novelty w = w #g,#r greater than k are allowed for each combination of values for the functions #g and #r (instead of 1).

1. ...