Added new reward method, and tried different training vars

.gitignore

@@ -132,4 +132,8 @@ dmypy.json
 
 # PyCharm
 .idea/
-__pycache__/
+__pycache__/
+
+models
+./models/
+models/

cube/helper_cube.py

@@ -86,14 +86,22 @@ def step(self, action):
     #
     #     return reward + solved_face
 
-    def reward_action(self, state, next_state):
-        reward = self.reward_color_count(next_state) - self.reward_color_count(state)
+    # def reward_action(self, state, next_state):
+    #     reward = self.reward_color_count(next_state) - self.reward_color_count(state)
+    #
+    #     if (self.reward_face_solved(state, next_state)) > 0:
+    #         reward = max(reward + 0.4, 1)
+    #
+    #     if self.check_solved():
+    #         reward = 1
+    #
+    #     return reward
 
-        if (self.reward_face_solved(state, next_state) / 6) > 0:
-            reward = max(reward + 0.4, 1)
+    def reward_action(self, state, next_state):
+        reward = CubeHelper.reward_face_solved(state, next_state) / 6
 
         if self.check_solved():
-            reward = 1
+            reward = 10
 
         return reward
 

solver/dqn_agent.py → dqn/dqn_agent.py

@@ -34,8 +34,6 @@ def __init__(self):
         self.rotation_dict = {0: "U", 1: "U'", 2: "D", 3: "D'", 4: "L", 5: "L'",
                               6: "R", 7: "R'", 8: "F", 9: "F'", 10: "B", 11: "B'"}
 
-        # self.prev_pred = None
-
     def create_model(self):
         """
         Creates the model for the neural network

solver/main.py → dqn/main.py

@@ -1,6 +1,6 @@
 from cube.helper_cube import CubeHelper
-from solver.dqn_agent import DqnAgent
-from solver.replay_buffer import ReplayBuffer
+from dqn.dqn_agent import DqnAgent
+from dqn.replay_buffer import ReplayBuffer
 
 import copy
 
@@ -16,6 +16,8 @@ def __init__(self):
         self.TARGET_UPDATE = 5
         self.UPDATE_ALL_TD = 2
 
+        self.model_save_path = "../models/model.h5"
+
     def train_model(self):
         for step in range(self.STEPS):
             self.get_train_data()
@@ -29,8 +31,9 @@ def train_model(self):
 
             if step % self.TARGET_UPDATE == 0:
                 self.agent.update_target_model()
+                self.agent.model.save(self.model_save_path)
 
-        self.agent.model.save("../models/model1.h5")
+        self.agent.model.save(self.model_save_path)
 
     def get_train_data(self):
         self.cube.scramble()

solver/test_predict.py → dqn/test_predict.py

@@ -6,10 +6,9 @@
 
 cube = CubeHelper()
 agent = DqnAgent()
-model = load_model('../models/model1.h5')
+model = load_model('../models/model.h5')
 
 agent.model = model
-[]
 def try_solve():
     cube.scramble()
     state = cube.get_cube_state()