4 ms·
Great article, have to wonder if somebody should great a version of alpha go zero as a "control". Then compare the level of effort in implementation and result
by koblas 8y ago
Great article, have to wonder if somebody should great a version of alpha go zero as a "control". Then compare the level of effort in implementation and results.
- mzl 8y agoThanks! It would really be interesting to see how AlphaZero style game play would work out. In particular, the very wide branching factor (when accounting for stochasticity) and relatively shallow game tree might change the characteristics noticeably. One thing that might be hard is to construct a reasonable game state representation for the neural network that is efficient for learning.
- koblas 8y agoThat's why it would be a fun comparison!