policytree

policytree copied to clipboard

Add penalized_policy_tree(X, Gamma1, Gamma2, penalty.type = c("sum", "ratio"), lambda = 1, etc...)

$\Gamma_{2,i}$ = 0 in a) is what policy_tree does, this PR just extends the "reward" calculation to compute a) or b), or the original one (just sum Gamma1). Since PT er very perf sensitive the a), b) or original calculation is resolved at compile-time.

perf.R

todos:

• timing
• more tests
• add penalized timing bench
• "seealso/readme"