Feat add total visits

cpp/command/gtp.cpp

@@ -826,6 +826,12 @@ struct GTPEngine {
         }
 
         const Board board = search->getRootBoard();
+
+        int totalVisits = 0;
+        for(int i = 0; i<buf.size(); i++) {
+          totalVisits += buf[i].numVisits;
+        }
+
         for(int i = 0; i<buf.size(); i++) {
           if(i > 0)
             out << " ";
@@ -849,6 +855,7 @@ struct GTPEngine {
           out << "info";
           out << " move " << Location::toString(data.move,board);
           out << " visits " << data.numVisits;
+          out << " totalVisits " << totalVisits;
           out << " utility " << utility;
           out << " winrate " << winrate;
           // We report lead for scoreMean here so that a bunch of legacy tools that use KataGo use lead instead, which

docs/GTP_Extensions.md

@@ -115,11 +115,12 @@ In addition to a basic set of [GTP commands](https://www.lysator.liu.se/~gunnar/
          * `pvVisits true` - Output the number of visits spent in the position after each move in each principal variation.
          * `pvEdgeVisits true` - Output the number of visits spent following each move in each principal variation.
       * Output format:
-         * Outputted lines look like `info move E4 visits 487 utility -0.0408357 winrate 0.480018 scoreMean -0.611848 scoreStdev 24.7058 scoreLead -0.611848 scoreSelfplay -0.515178 prior 0.221121 lcb 0.477221 utilityLcb -0.0486664 order 0 pv E4 E3 F3 D3 F4 P4 P3 O3 Q3 O4 K3 Q6 S6 E16 E17 info move P16 visits 470 utility -0.0414945 winrate 0.479712 scoreMean -0.63075 scoreStdev 24.7179 scoreLead -0.63075 scoreSelfplay -0.5221 prior 0.220566 lcb 0.47657 utilityLcb -0.0502929 order 1 pv P16 P17 O17 Q17 O16 E17 H17 D15 C15 D14 C13 D13 C12 D12 info move E16 visits 143 utility -0.0534071 winrate 0.474509 scoreMean -0.729858 scoreStdev 24.7991 scoreLead -0.729858 scoreSelfplay -0.735747 prior 0.104652 lcb 0.470674 utilityLcb -0.0641425 order 2 pv E16 P4 P3 O3 Q3 O4 E3 H3 D5 C5`
+         * Outputted lines look like `info move E4 visits 487 totalVisits 1024 utility -0.0408357 winrate 0.480018 scoreMean -0.611848 scoreStdev 24.7058 scoreLead -0.611848 scoreSelfplay -0.515178 prior 0.221121 lcb 0.477221 utilityLcb -0.0486664 order 0 pv E4 E3 F3 D3 F4 P4 P3 O3 Q3 O4 K3 Q6 S6 E16 E17 info move P16 visits 470 utility -0.0414945 winrate 0.479712 scoreMean -0.63075 scoreStdev 24.7179 scoreLead -0.63075 scoreSelfplay -0.5221 prior 0.220566 lcb 0.47657 utilityLcb -0.0502929 order 1 pv P16 P17 O17 Q17 O16 E17 H17 D15 C15 D14 C13 D13 C12 D12 info move E16 visits 143 utility -0.0534071 winrate 0.474509 scoreMean -0.729858 scoreStdev 24.7991 scoreLead -0.729858 scoreSelfplay -0.735747 prior 0.104652 lcb 0.470674 utilityLcb -0.0641425 order 2 pv E16 P4 P3 O3 Q3 O4 E3 H3 D5 C5`
          * `info` - Indicates the start of information for a new possible move, followed by key-value pairs. Current key-value pairs:
             * **NOTE: Consumers of this data should attempt to be robust to the order of these fields, as well as to possible addition of new fields in the future.**
             * `move` - The move being analyzed.
             * `visits` - The number of visits invested into the move so far.
+            * `totalVisits` - The number of total visits of all the moves so far. The `totalVisits` sums up all the `visits` value of all the moves in the current output.
             * `winrate` - The winrate of the move so far, as a float in [0,1].
             * `scoreMean` - Same as scoreLead. "Mean" is a slight misnomer, but this field exists to preserve compatibility with existing tools.
             * `scoreStdev` - The predicted standard deviation of the final score of the game after this move, in points. (NOTE: due to the mechanics of MCTS, this value will be **significantly biased high** currently, although it can still be informative as a *relative* indicator).