added passing of main_config and train_config to rl_training.py

engine/src/rl/rl_loop.py

@@ -143,6 +143,9 @@ def __init__(self, args, nb_games_to_update=1024, nb_arena_games=100, lr_reducti
         binary_mappings = {"crazyhouse": "CrazyAra",
                 "chess": "ClassicAra"}
         self.crazyara_binary_name = binary_mappings[self.args.uci_variant]
+        main_config["planes_train_dir"] =  args.crazyara_binary_dir + "export/train/"
+        main_config["planes_val_dir"] =  args.crazyara_binary_dir + "export/val/"
+
         self.proc = None
         self.nb_games_to_update = nb_games_to_update
         if nb_arena_games % 2 == 1:
@@ -242,7 +245,9 @@ def _set_uci_options(self, is_arena=False):
         # set_uci_param(self.proc, "MaxInitPly", 30)
         set_uci_param(self.proc, "Reuse_Tree", "false")
         set_uci_param(self.proc, "Precision", self.args.precision)
-        # set_uci_param(self.proc, "Selfplay_Number_Chunks", 1)
+        #set_uci_param(self.proc, "Selfplay_Number_Chunks", 1)
+        set_uci_param(self.proc, "Selfplay_Number_Chunks", 640)
+        set_uci_param(self.proc, "Selfplay_Chunk_Size", 128)
 
         if is_arena is True:
 #            set_uci_param(self.proc, "Centi_Temperature", 60) cz
@@ -490,7 +495,8 @@ def check_for_enough_train_data(self, number_files_to_update):
             process = Process(target=update_network, args=(queue, self.nn_update_index, self.k_steps,
                                                            self.max_lr, self._get_current_model_arch_file(),
                                                            self._get_current_model_weight_file(),
-                                                           self.crazyara_binary_dir, not self.args.no_onnx_export))
+                                                           self.crazyara_binary_dir, not self.args.no_onnx_export,
+                                                           main_config, train_config))
             logging.info("start training")
             process.start()
             self.k_steps = queue.get() + 1
@@ -551,7 +557,7 @@ def parse_args(cmd_args: list):
     parser.add_argument('--nn-update-idx', type=int, default=0,
                         help="Index of how many NN updates have been done so far."
                              " This will be used to label the NN weights (default: 0)")
-    parser.add_argument("--nn-update-files", type=int, default=15, #10,
+    parser.add_argument("--nn-update-files", type=int, default=10,
                         help="How many new generated training files are needed to apply an update to the NN")
     parser.add_argument("--arena-games", type=int, default=100,
                         help="How many arena games will be done to judge the quality of the new network")

engine/src/rl/rl_training.py

@@ -13,8 +13,6 @@
 import mxnet as mx
 
 sys.path.append("../../../")
-from DeepCrazyhouse.configs.main_config import main_config
-from DeepCrazyhouse.configs.train_config import train_config
 from DeepCrazyhouse.src.preprocessing.dataset_loader import load_pgn_dataset
 from DeepCrazyhouse.src.training.trainer_agent import acc_sign, cross_entropy, acc_distribution
 from DeepCrazyhouse.src.training.trainer_agent_mxnet import TrainerAgentMXNET, add_non_sparse_cross_entropy,\
@@ -24,7 +22,7 @@
 from DeepCrazyhouse.src.domain.neural_net.onnx.convert_to_onnx import convert_mxnet_model_to_onnx
 
 
-def update_network(queue, nn_update_idx, k_steps_initial, max_lr, symbol_filename, params_filename, cwd, convert_to_onnx):
+def update_network(queue, nn_update_idx, k_steps_initial, max_lr, symbol_filename, params_filename, cwd, convert_to_onnx, main_config, train_config):
     """
     Creates a new NN checkpoint in the model contender directory after training using the game files stored in the
      training directory
@@ -38,6 +36,8 @@ def update_network(queue, nn_update_idx, k_steps_initial, max_lr, symbol_filenam
     Updates the neural network with the newly acquired games from the replay memory
     :param cwd: Current working directory (must end with "/")
     :param convert_to_onnx: Boolean indicating if the network shall be exported to ONNX to allow TensorRT inference
+    :param main_config: Dict of the main_config (imported from main_config.py)
+    :param train_config: Dict of the train_config (imported from train_config.py)
     :return: k_steps_final
     """