DT's reward-to-go vs. QL's Bellman incl. discount, not choice of architecture for policy. You could also do DTs with RNNs (though own problems w/ memory). Apol

DT's reward-to-go vs. QL's Bellman incl. discount, not choice of architecture for policy. You could also do DTs with RNNs (though own problems w/ memory). Apologies if we're talking past one another.