Added credentials, log for number of games used in training, and upda…

…ted parameters

GoCoach.py

@@ -104,10 +104,6 @@ def learn(self):
         if self.args.load_model:
             self.loadLosses()
 
-        print(self.p_loss_per_iteration)
-        print(self.v_loss_per_iteration)
-        print(self.winRate)
-
         for i in range(self.args.start_iter, self.args.numIters + 1):
             iterHistory['ITER'].append(i)
             # bookkeeping
@@ -176,11 +172,19 @@ def learn(self):
                             append_downloads = True
 
                 downloads_count = downloads_count * 5
-                downloads_count += 100
+                downloads_count += self.args.numEps
             else:
-                downloads_count = self.args.numEps                
+                downloads_count = self.args.numEps     
+
+            #Log how many games were added during each iteration
+            file_name = f'logs/go/{self.args.nettype}_MCTS_SimModified_checkpoint/{self.args.boardsize}/Game_Counts.txt'
+            if not os.path.isfile(file_name):
+                counts_file = open(file_name, 'w')
+                counts_file.close()
+            counts_file = open(file_name, 'a')
+            counts_file.write(f"\n Number of games added to train examples during iteration #{i}: {downloads_count} games\n")
+            counts_file.close()
 
-            print(len(self.iterationTrainExamples))
             # save the iteration examples to the history
             if not self.skipFirstSelfPlay or append_downloads:
                 self.trainExamplesHistory.append(self.iterationTrainExamples)
@@ -225,6 +229,10 @@ def learn(self):
                 print('REJECTING NEW MODEL')
                 iterHistory['PITT_RESULT'].append('R')
                 self.nnet.load_checkpoint(folder=self.args.checkpoint, filename='temp.pth.tar')
+                if i == 1 and self.args.distributed_training and not self.args.load_model:
+                    upload_path = os.path.join(self.args.checkpoint, 'temp.pth.tar')
+                    drive.FileUpload(upload_path, upload_number)
+
             else:
                 print('ACCEPTING NEW MODEL')
                 iterHistory['PITT_RESULT'].append('A')

GoMCTS.py

@@ -51,7 +51,7 @@ def getActionProb(self, canonicalBoard, temp=1):
         # display(canonicalBoard)
 
         #print('current sim numbers:{}'.format(max(self.args.numMCTSSims,self.smartSimNum)))
-        for i in range(max(self.args.numMCTSSims,self.smartSimNum)):
+        for i in range(min(self.args.numMCTSSims,self.smartSimNum)):
 
             self.search(canonicalBoard)
 

credentials.json

@@ -0,0 +1 @@
+{"installed":{"client_id":"778638443708-f54vki1i89tlc8jd4kemfodbu8c0gca9.apps.googleusercontent.com","project_id":"ai2go-391221","auth_uri":"https://accounts.google.com/o/oauth2/auth","token_uri":"https://oauth2.googleapis.com/token","auth_provider_x509_cert_url":"https://www.googleapis.com/oauth2/v1/certs","client_secret":"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","redirect_uris":["http://localhost"]}}

go/pytorch/NNet.py

@@ -28,7 +28,7 @@
 
 args = dotdict({
     'lr': 0.001,
-    'dropout': 0.3,
+    'dropout': 0.0,
     'epochs': 10,
     'cuda': torch.cuda.is_available(),
     'num_channels': 512,

main.py

@@ -22,25 +22,25 @@ class Display(IntEnum):
 
 
 args = dotdict({
-    'numIters': 5,
-    'numEps': 2,              # Number of complete self-play games to simulate during a new iteration.
+    'numIters': 1000,
+    'numEps': 100,              # Number of complete self-play games to simulate during a new iteration.
     'tempThreshold': 15,
     'updateThreshold': 0.54,    # During arena playoff, new neural net will be accepted if threshold or more of games are won.
     'maxlenOfQueue': 200000,    # Number of game examples to train the neural networks.
     'numMCTSSims': 150,         # Number of games moves for MCTS to simulate.
-    'arenaCompare': 1,         # Number of games to play during arena play to determine if new net will be accepted.
-    'cpuct': 3,
+    'arenaCompare': 50,         # Number of games to play during arena play to determine if new net will be accepted.
+    'cpuct': 1.0,
 
     'checkpoint': './logs/go/{}_checkpoint/{}/'.format(NetType + '_' + tag, BoardSize),
-    'load_model': True,
+    'load_model': False,
     'numItersForTrainExamplesHistory': 25,
     'display': Display.DISPLAY_BOARD,
     'datetime': datetime.now().strftime("%d-%m-%Y %H:%M"),
     'sgf_datetime': datetime.now().strftime("%d-%m-%Y %H"),
     'nettype': NetType,
     'boardsize': BoardSize,
     'distributed_training': True,
-    'start_time': time.time()
+    'start_time': time.time(),
 })
 if args.load_model:
     checkpoint_dir = f'logs/go/{NetType}_MCTS_SimModified_checkpoint/{BoardSize}/'