refactored files

machine_player.py

@@ -0,0 +1,113 @@
+import numpy as np
+from a_player import APlayer
+from helper import sample
+from numpy.random import uniform
+
+
+class MachinePlayer(APlayer):
+    def __init__(self, mark='X', with_debug=False, values=None):
+        APlayer.__init__(self, mark)
+        self.p_exp = 0.2
+        self.alpha = 0.1
+        self.with_debug = with_debug
+        if values is None:
+            self.init_values()
+        else:
+            self.values = values
+
+    def init_values(self):
+        self.values = {}
+
+    def print_state_table(self):
+        count = 0
+        for key in self.values:
+            print(f"{self.mark}: {count:05d}: {key} = {self.values[key]:0.3f}")
+            count += 1
+
+    @staticmethod
+    def to_key(state: np.array):
+        key = ''
+        for st in state.reshape(state.size):
+            key += st
+        return key
+
+    def get_value(self, state: np.array):
+        key = self.to_key(state)
+        try:
+            result = self.values[key]
+        except KeyError:
+            result = 0.5
+
+        return result
+
+    def set_value(self, state: np.array, value):
+        self.values[self.to_key(state)] = value
+
+    def reward(self, value):
+        self.set_value(self.state, value)
+
+    @staticmethod
+    def get_potential_moves(state: np.array) -> np.array:
+        st = state.reshape(state.size)
+        indices = [idx for idx, s in enumerate(st) if '-' in s]
+        return np.array(indices)
+
+    def new_game(self):
+        self.state = None
+        self.state_last = None
+
+    def move(self, state: np.array):
+        do_sample = False
+        values = np.array([])
+        # get possible move
+        moves = self.get_potential_moves(state)
+        can_move = moves.size > 0
+        best_move = None
+        best_value = -1
+        if can_move:
+            for move in moves:
+                # create hypothetical next state
+                state_next = self.state_from_move(state.copy(), move)
+                # evaluate value
+                value = self.get_value(state_next)
+                if best_value < value:
+                    best_value = value
+                    best_move = move
+                values = np.append(values, value)
+
+            next_move = best_move
+            is_exp = False
+            if uniform() < self.p_exp:
+                is_exp = True
+                index = np.random.randint(len(moves))
+                next_move = moves[index]
+            elif do_sample:
+                index = sample(values)
+                next_move = moves[index]
+
+            if self.with_debug:
+                print(f"{self.mark}: Values    = {values}")
+                print(f"{self.mark}: Moves     = {moves+1}")
+                print(f"{self.mark}: Best move = {best_move+1}")
+                print(f"{self.mark}: Next move = {next_move+1}, is_exp={is_exp}")
+
+            if self.state is not None:
+                self.state_last = self.state.copy()
+
+            self.state = self.state_from_move(state.copy(), next_move)
+
+            # Learn
+            if not is_exp and self.state_last is not None:
+                v0 = self.get_value(self.state_last)
+                v1 = self.get_value(self.state)
+                d = v0 + self.alpha*(v1-v0)
+                if d > 0:
+                    self.set_value(self.state_last, d)
+                if self.with_debug:
+                    print(f"{self.mark}: Learned {d:0.3f}")
+
+        return self.state, can_move
+
+    def state_from_move(self, state: np.array, field):
+        state.reshape(state.size)[field] = self.mark
+        return state