Interpreting learned search: finding a transition model and value function in an RNN that plays Sokoban

Re-assign community