Paste code

private void runRAVE() {
        GameFactory factory = gameFactory.copy();
        ArrayList<Node> visited = new ArrayList<>();
        ArrayList<AMAFNode> siblingsOfVisited = new ArrayList<>();
        ArrayList<int[]> sourceActionsTakenByDecisionPlayer = new ArrayList<>();
        int decisionPlayer = tree.getRoot().getPlayer();
        Node currentNode = tree.getRoot();
        visited.add(currentNode);
        int depth = 1;
        while(!currentNode.isLeaf() && !currentNode.isTerminal() && currentNode.visited() && depth < mctsConfig.maxTreeDepth){
            Node parentNode = currentNode;
            if (currentNode.isChanceNode()){
                Game game = factory.getGame(currentNode.getState());
                game.gameTick(true);
                currentNode = new Node(game.getState(), game.isTerminal(), game.getCurrentPlayer());
                tree.putNodeIfAbsent(currentNode);
            } else {
                currentNode = RAVE.selectChild(currentNode, tree, mctsConfig.C, mctsConfig.minVisits, mctsConfig.k);
                visited.add(currentNode);
                ArrayList<Node> siblingsOfCurrentNode = parentNode.getChildren(tree);

                if (parentNode.getPlayer() == decisionPlayer) {
                    sourceActionsTakenByDecisionPlayer.add(currentNode.getSourceAction());
                    for (Node child : siblingsOfCurrentNode) {
                        AMAFNode amafNode = (AMAFNode) child;
                        if (!Arrays.equals(currentNode.getState(), amafNode.getState()))
                            siblingsOfVisited.add(amafNode);
                    }
                }
            }
            depth++;
        }
        if (currentNode.isLeaf() && currentNode.visited() && !currentNode.isTerminal()){
            Node randomExpandedNode = tree.expand(currentNode, factory, mctsConfig.selectionPolicy);
            Node parentNode = tree.getTreeNode(new NodeState(currentNode.getState(), currentNode.getSourceAction()));
            ArrayList<Node> siblingsOfExpandedNode = parentNode.getChildren(tree);

            if (parentNode.getPlayer() == decisionPlayer) {
                sourceActionsTakenByDecisionPlayer.add(randomExpandedNode.getSourceAction());
                for (Node child : siblingsOfExpandedNode) {
                    AMAFNode amafNode = (AMAFNode) child;
                    if (!Arrays.equals(randomExpandedNode.getState(), amafNode.getState()))
                        siblingsOfVisited.add(amafNode);
                }
            }
            visited.add(randomExpandedNode);
            currentNode = randomExpandedNode;
        } else {
            // currentNode not visited enough times, proceed with rollout
        }
        Game game = factory.getGame(currentNode.getState());
        double[] wins = new double[GameFactory.nMaxPlayers()];
        for (int i = 0; i < mctsConfig.nRolloutsPerIteration; i++) {
            Game gameClone = game.copy();
            while(!gameClone.isTerminal()){
                int[] action = gameClone.getRandomLegalAction();
                if (gameClone.getCurrentPlayer() == decisionPlayer)
                    sourceActionsTakenByDecisionPlayer.add(action);
                gameClone.performAction(action, true);
            }
            if (gameClone.getWinner() != -1) {
                wins[gameClone.getWinner()] += 1.0;
            } else {
                for (int pl = 0; pl < wins.length; pl++)
                    wins[pl] = 0.5;
            }
        }
        final int nRollouts = mctsConfig.nRolloutsPerIteration;
        HashSet<Node> v = new HashSet<>(visited.size());
        v.addAll(visited);
        v.forEach(n -> n.update(wins, nRollouts));
        v.forEach(n -> n.updateVisitedStatus(mctsConfig.minVisits));
        for (int j = 0; j < sourceActionsTakenByDecisionPlayer.size(); j++){
            int[] actionEncounteredByDecisionPlayer = sourceActionsTakenByDecisionPlayer.get(j);
            for (int k = 0; k < siblingsOfVisited.size(); k++){
                AMAFNode sibling = siblingsOfVisited.get(k);
                if (Arrays.equals(actionEncounteredByDecisionPlayer, sibling.getSourceAction())){
                    sibling.updateAMAF(wins, nRollouts);
                    // TicTacToe3D analysis:
                    // if Action was good for decision player 1 and &#039;contributed&#039; to a win, wins: [0, 1, 0}
                    // When we decide between amafNode and other nodes, we want to see how many wins its action brought US - the selectionArgNode.getPlayer
                    // if we are decisionPlayer we want to access wins[1] which is wins[decisionPlayer] = wins[root.getCurrentPlayer()]
                    // if we analyze the node of the opponent we want to access wins[2] which is wins[node.getCurrentPlayer()] in the selection
                }
            }
        }
    }