Adapt the code to train on Three map

There was some changes on getting the Region information to generate the State representation for Protobuf.

bandana/src/main/java/cruz/agents/DeepDip.java

@@ -16,6 +16,7 @@ public class DeepDip extends DumbBot {
     // The OpenAI Adapter contains the necessary functions and fields to make the connection to the Open AI environment
     private OpenAIAdapter openAIAdapter;
     private Logger logger = new Logger();
+    private String winner = null;
 
     private DeepDip(String name, int finalYear, String logPath) {
         super(name, finalYear, logPath);
@@ -195,6 +196,7 @@ private List<MTOOrder> getMTOOrders(List<Order> orders) {
 
     @Override
     public void handleSlo(String winner) {
+        this.winner = winner;
         this.openAIAdapter.setWinner(winner);
         if (this.me.getName().equals(winner)) {
             System.out.println("GAME RESULT: " + this.me + " won with a solo victory.");
@@ -207,10 +209,17 @@ public void handleSlo(String winner) {
     @Override
     public void handleSMR(String[] message) {
         GameResult gameResult = new GameResult(message, 2);
+
+        if (this.me.getControlledRegions().size() < 1) {
+            this.openAIAdapter.setWinner("eliminated");
+        } else if (this.winner == null) {
+            this.openAIAdapter.setWinner("draw");
+        }
         this.openAIAdapter.endOfGame(gameResult);
 
         System.out.println("END GAME: " + Arrays.toString(message));
         super.handleSMR(message);
+        this.winner = null;
     }
 
     public Logger getLogger() {

bandana/src/main/java/cruz/agents/OpenAIAdapter.java

@@ -29,7 +29,7 @@ public class OpenAIAdapter {
     public static final int INVALID_DEAL_REWARD = -10;
 
     /** Reward given for winning the game. */
-    public static final int SC_WIN = 5;
+    public static final int SC_WIN = 8;
 
     /** The OpenAINegotiator instance to which this adapter is connected. */
     public OpenAINegotiator agent;
@@ -184,46 +184,48 @@ private void generatePowerNameToIntMap() {
         }
     }
 
+    /**
+     * First process all Provinces.
+     * Then add the Owners and Units of each Province.
+     * Then add the created Provinces to the Observation.
+     */
     private ProtoMessage.ObservationData generateObservationData() {
         ProtoMessage.ObservationData.Builder observationDataBuilder = ProtoMessage.ObservationData.newBuilder();
         Map<String, ProtoMessage.ProvinceData.Builder> nameToProvinceDataBuilder = new HashMap<>();
 
         String agent_name = (this.agent2 == null)? this.agent.me.getName() : this.agent2.getMe().getName();
         observationDataBuilder.setPlayer(powerNameToInt.get(agent_name));
 
-        // FIRST PROCESS ALL PROVINCES
-        Vector<Province> provinces = (this.agent2 == null) ? this.agent.game.getProvinces() : this.agent2.getGame().getProvinces();
+        Vector<Region> regions = (this.agent2 == null) ? this.agent.game.getRegions() : this.agent2.getGame().getRegions();
         int id = 1;
-        for (Province p : provinces) {
+        for (Region r : regions) {
             ProtoMessage.ProvinceData.Builder provinceDataBuilder = ProtoMessage.ProvinceData.newBuilder();
-            int isSc = p.isSC() ? 1 : 0;
+            int isSc = r.getProvince().isSC() ? 1 : 0;
 
             provinceDataBuilder.setId(id);
             provinceDataBuilder.setSc(isSc);
 
-            nameToProvinceDataBuilder.put(p.getName(), provinceDataBuilder);
+            nameToProvinceDataBuilder.put(r.getName(), provinceDataBuilder);
 
             id++;
         }
 
-        // THEN ADD THE OWNERS & UNITS OF EACH PROVINCE
         List<Power> powers = (this.agent2 == null)? this.agent.game.getPowers():this.agent2.getGame().getPowers();
         for (Power pow : powers) {
             for (Province p : pow.getOwnedSCs()) {
-                // Get the correspondent province builder and add the current owner of the province
-                ProtoMessage.ProvinceData.Builder provinceDataBuilder = nameToProvinceDataBuilder.get(p.getName());
-                provinceDataBuilder.setOwner(powerNameToInt.get(pow.getName()));
+                for (Region r : p.getRegions()) {
+                    ProtoMessage.ProvinceData.Builder provinceDataBuilder = nameToProvinceDataBuilder.get(r.getName());
+                    provinceDataBuilder.setOwner(powerNameToInt.get(pow.getName()));
+                }
             }
 
             for (Region r : pow.getControlledRegions()) {
-                Province p = r.getProvince();
-                ProtoMessage.ProvinceData.Builder provinceDataBuilder = nameToProvinceDataBuilder.get(p.getName());
+                ProtoMessage.ProvinceData.Builder provinceDataBuilder = nameToProvinceDataBuilder.get(r.getName());
                 provinceDataBuilder.setOwner(powerNameToInt.get(pow.getName()));
                 provinceDataBuilder.setUnit(powerNameToInt.get(pow.getName()));
             }
         }
 
-        // ADD CREATED PROVINCES TO OBSERVATION
         for (Map.Entry<String, ProtoMessage.ProvinceData.Builder> entry : nameToProvinceDataBuilder.entrySet()) {
             observationDataBuilder.addProvinces(entry.getValue().build());
         }
@@ -247,7 +249,7 @@ private void rewardFunction() {
             String agent_name = (this.agent2 == null)? this.agent.me.getName() : this.agent2.getMe().getName();
             if (agent_name.equals(this.winner)) {
                 this.reward += SC_WIN;
-            } else {
+            } else if (!this.winner.equals("draw")){
                 this.reward -= SC_WIN;
             }
         } else {
@@ -259,8 +261,6 @@ private BasicDeal generateDeal(ProtoMessage.DealData dealData) {
         List<DMZ> dmzs = new ArrayList<>();
         List<OrderCommitment> ocs = new ArrayList<>();
 
-
-        // Add MY order commitment
         Province ourStartProvince = this.agent.game.getProvinces().get(dealData.getOurMove().getStartProvince());
         Province ourDestinationProvince = this.agent.game.getProvinces().get(dealData.getOurMove().getDestinationProvince());
 
@@ -303,24 +303,14 @@ private BasicDeal generateDeal(ProtoMessage.DealData dealData) {
     private List<Order> generateOrders(ProtoMessage.OrdersData ordersData) {
         List<Order> orders = new ArrayList<>();
         List<ProtoMessage.OrderData> support_orders = new ArrayList<>();
-        List<Province> game_provinces = this.agent2.getGame().getProvinces();
         List<Region> game_regions = this.agent2.getGame().getRegions();
 
         for (ProtoMessage.OrderData order : ordersData.getOrdersList()) {
             if (order.getStart() == -1){
                 break;
             }
-            Province start_province = game_provinces.get(order.getStart());
-            Province destination_province = game_provinces.get(order.getDestination());
-
-            Region start = game_regions.stream()
-                .filter(r -> r.getProvince().getName().equals(start_province.getName()))
-                .findAny()
-                .orElse(null);
-            Region destination = game_regions.stream()
-                .filter(r -> r.getProvince().getName().equals(destination_province.getName()))
-                .findAny()
-                .orElse(null);
+            Region start = game_regions.get(order.getStart());
+            Region destination = game_regions.get(order.getDestination());
 
             if (order.getAction() == 0) {
                 orders.add(new HLDOrder(this.agent2.getMe(), start));
@@ -331,8 +321,6 @@ private List<Order> generateOrders(ProtoMessage.OrdersData ordersData) {
                     support_orders.add(order);
                 }
             } else {
-                //System.err.println("WRONG BORDER: For order of type " + order.getAction() + ", the destination " + destination + " is not a border with current province " + start);
-                //this.addReward(INVALID_DEAL_REWARD);
                 orders.add(new HLDOrder(this.agent2.getMe(), start));
             }
         }
@@ -345,8 +333,6 @@ private List<Order> generateOrders(ProtoMessage.OrdersData ordersData) {
                 .findAny()
                 .orElse(null);
             if (order_to_support == null) {
-                //System.err.println("ORDER TO SUPPORT NOT FOUND");
-                //this.addReward(INVALID_DEAL_REWARD);
                 orders.add(new HLDOrder(this.agent2.getMe(), start));
             } else if (order_to_support instanceof MTOOrder) {
                 orders.add(new SUPMTOOrder(this.agent2.getMe(), start, (MTOOrder) order_to_support));

bandana/src/main/java/ddejonge/bandana/tournament/TournamentRunner.java

@@ -10,13 +10,14 @@
 public class TournamentRunner {
     final static boolean MODE = false;  //Strategy/false vs Negotiation/true
 	final static int REMOTE_DEBUG = 0;	//Set whether I want to remote debug the OpenAI jar or not
-    private final static String GAME_MAP = "small"; //Game map can be 'standard', 'mini' or 'small'
+    private final static String GAME_MAP = "three"; //Game map can be 'standard', or 'mini', or 'small', or 'three'
     private final static String FINAL_YEAR = "2000"; //The year after which the agents in each game are supposed to propose a draw to each other.
 
     // Using a custom map to define how many players are there on each custom map
     private final static Map<String, Integer> mapToNumberOfPlayers  = new HashMap<String, Integer>() {{
         put("standard", 7);
         put("small", 2);
+        put("three", 3);
     }};
 
 	// Main folder where all the logs are stored. For each tournament a new folder will be created inside this folder
@@ -79,13 +80,12 @@ public static void run(int numberOfGames, int moveTimeLimit, int retreatTimeLimi
             //Create a list of ScoreCalculators to determine how the players should be ranked in the tournament.
             ArrayList<ScoreCalculator> scoreCalculators = new ArrayList<ScoreCalculator>();
 
-            if(GAME_MAP.toLowerCase().equals("standard")) {
+            if (GAME_MAP.toLowerCase().equals("standard")) {
                 scoreCalculators.add(new SoloVictoryCalculator());
                 scoreCalculators.add(new SupplyCenterCalculator());
                 scoreCalculators.add(new PointsCalculator());
                 scoreCalculators.add(new RankCalculator());
-            }
-            else {
+            } else {
                 scoreCalculators.add(new RankCalculator());
             }
 
@@ -109,7 +109,7 @@ public static void run(int numberOfGames, int moveTimeLimit, int retreatTimeLimi
                         name = "DeepDip";
                         command = deepDipCommand;
                     } else {
-                        name = "DumbBot";
+                        name = "DumbBot " + i;
                         command = dumbBotCommand;
                     }
 
@@ -159,7 +159,7 @@ public static void run(int numberOfGames, int moveTimeLimit, int retreatTimeLimi
                 playerProcess.destroy();
             }
 
-            if(tournamentObserver != null){
+            if (tournamentObserver != null) {
                 tournamentObserver.exit();
             }
 

gym-diplomacy/gym_diplomacy/envs/diplomacy_strategy_env.py

@@ -25,8 +25,12 @@
 
 ### CONSTANTS
 NUMBER_OF_ACTIONS = 3
-NUMBER_OF_PLAYERS = 2
-NUMBER_OF_PROVINCES = 19#8#75
+MAPS = ['mini', 'small', 'three', 'standard']
+CURRENT_MAP = MAPS[2]
+PLAYERS = {'mini':2, 'small':2, 'three':3, 'standard':7}
+NUMBER_OF_PLAYERS = PLAYERS[CURRENT_MAP]
+REGIONS = {'mini':10, 'small':19, 'three':37, 'standard':121}
+NUMBER_OF_PROVINCES = REGIONS[CURRENT_MAP]
 
 
 def observation_data_to_observation(observation_data: proto_message_pb2.ObservationData) -> np.array:
@@ -58,6 +62,9 @@ def observation_data_to_observation(observation_data: proto_message_pb2.Observat
     done = observation_data.done
     info = {}
 
+    if len(get_player_units(observation)) < 1:
+        done = True
+
     return observation, reward, done, info