3 files changed, 426 insertions, 0 deletions
diff --git a/dotsandboxes/agents/algorithms/MCTS.py b/dotsandboxes/agents/algorithms/MCTS.py
new file mode 100644
index 0000000..6c71ba9
--- /dev/null
+++ b/dotsandboxes/agents/algorithms/MCTS.py
@@ -0,0 +1,151 @@
+import math
+from copy import deepcopy
+from time import perf_counter
+from random import choice
+
+from GameState import GameState
+
+# Based on https://github.com/DieterBuys/mcts-player/
+
+class GameController(object):
+    def get_next_move(self, state):
+        # when you get a new move, it is assumed that the game is not ended yet
+        assert state.get_moves()
+
+
+class MCTSNode(object):
+    """Monte Carlo Tree Node.
+    Each node encapsulates a particular game state, the moves that
+    are possible from that state and the strategic information accumulated
+    by the tree search as it progressively samples the game space.
+    """
+
+    def __init__(self, state, parent=None, move=None):
+        self.parent = parent
+        self.move = move
+        self.state = state
+
+        self.plays = 0
+        self.score = 0
+
+        self.pending_moves = state.get_moves()
+        self.children = []
+
+    def select_child_ucb(self):
+        # Note that each node's plays count is equal
+        # to the sum of its children's plays
+        def ucb(child):
+            win_ratio = child.score / child.plays \
+                        + math.sqrt(2 * math.log(self.plays) / child.plays)
+            return win_ratio
+
+        return max(self.children, key=ucb)
+
+    def expand_move(self, move):
+        self.pending_moves.remove(move)  # raises KeyError
+
+        child_state = deepcopy(self.state)
+        child_state.play_move(move)
+
+        child = MCTSNode(state=child_state, parent=self, move=move)
+        self.children.append(child)
+        return child
+
+    def get_score(self, result):
+        # return result
+        if result == 0.5:
+            return result
+
+        if self.state.player == 2:
+            if self.state.next_turn_player == result:
+                return 0.0
+            else:
+                return 1.0
+        else:
+            if self.state.next_turn_player == result:
+                return 1.0
+            else:
+                return 0.0
+
+        if self.state.next_turn_player == result:
+            return 0.0
+        else:
+            return 1.0
+
+    def __repr__(self):
+        s = 'ROOT\n' if self.parent is None else ''
+
+        children_moves = [c.move for c in self.children]
+
+        s += """Score ratio: {score} / {plays}
+Pending moves: {pending_moves}
+Children's moves: {children_moves}
+State:
+{state}\n""".format(children_moves=children_moves, **self.__dict__)
+
+        return s
+
+
+class MCTSGameController(GameController):
+    """Game controller that uses MCTS to determine the next move.
+    This is the class which implements the Monte Carlo Tree Search algorithm.
+    It builds a game tree of MCTSNodes and samples the game space until a set
+    time has elapsed.
+    """
+
+    def select(self):
+        node = self.root_node
+
+        # Descend until we find a node that has pending moves, or is terminal
+        while node.pending_moves == set() and node.children != []:
+            node = node.select_child_ucb()
+
+        return node
+
+    def expand(self, node):
+        assert node.pending_moves != set()
+
+        move = choice(tuple(node.pending_moves))
+        return node.expand_move(move)
+
+    def simulate(self, state, max_iterations=1000):
+        state = deepcopy(state)
+
+        move = state.get_random_move()
+        while move is not None:
+            state.play_move(move)
+            move = state.get_random_move()
+
+            max_iterations -= 1
+            if max_iterations <= 0:
+                return 0.5  # raise exception? (game too deep to simulate)
+
+        return state.game_result
+
+    def update(self, node, result):
+        while node is not None:
+            node.plays += 1
+            node.score += node.get_score(result)
+            node = node.parent
+
+    def get_next_move(self, state, time_allowed=1.0):
+        super(MCTSGameController, self).get_next_move(state)
+
+        # Create new tree (TODO: Preserve some state for better performance?)
+        self.root_node = MCTSNode(state)
+        iterations = 0
+
+        start_time = perf_counter()
+        while perf_counter() < start_time + time_allowed:
+            node = self.select()
+
+            if node.pending_moves != set():
+                node = self.expand(node)
+
+            result = self.simulate(node.state)
+            self.update(node, result)
+
+            iterations += 1
+
+        # Return most visited node's move
+        return max(self.root_node.children, key=lambda n: n.plays).move
diff --git a/dotsandboxes/agents/algorithms/alphaBeta.py b/dotsandboxes/agents/algorithms/alphaBeta.py
new file mode 100644
index 0000000..8e041fe
--- /dev/null
+++ b/dotsandboxes/agents/algorithms/alphaBeta.py
@@ -0,0 +1,105 @@
+from GameState import GameState
+class GameController(object):
+    def get_next_move(self, state):
+        # when you get a new move, it is assumed that the game is not ended yet
+        assert state.get_moves()
+
+
+def alpha_beta(node, alpha, beta):
+
+    # Based on https://en.wikipedia.org/wiki/Alpha%E2%80%93beta_pruning#Pseudocode
+    # node needs to support three operations: isTerminal(), value(), getChildren(), maximizingPlayer()
+
+    if node.isTerminal():
+        return node.value()
+
+    if node.maximizingPlayer():
+
+        v = float("-inf")
+        for child in node.getChildren():
+
+            v = max(v, alpha_beta(child, alpha, beta))
+            alpha = max(alpha, v)
+            if beta <= alpha:
+                break
+
+    else:
+
+        v = float("inf")
+        for child in node.getChildren():
+
+            v = min(v, alpha_beta(child, alpha, beta))
+            beta = min(beta, v)
+            if beta <= alpha:
+                break
+
+    return v
+
+
+# A class for defining algorithms used (minimax and alpha-beta pruning)
+class AlphaBeta:
+
+    def miniMax(State, Ply_num):  # Function for the minimax algorithm
+
+        for i in range(State.Current.dimY):
+            for j in range(State.Current.dimX):
+                if State.Current.Mat[i][j] == ' ' and (j, i) not in State.children:
+                    State.Make(j, i, True)
+                    if Ply_num < 2:
+                        return (i, j)
+
+        Minimum_Score = 1000
+        i = 0
+
+        j = 0
+        for k, z in State.children.items():
+            Result = Algo.Maximum(z, Ply_num - 1, Minimum_Score)
+            if Minimum_Score > Result:
+                Minimum_Score = Result
+                i = k[0]
+                j = k[1]
+
+        return (i, j)
+
+    # Alpha-beta pruning function for taking care of Alpha values
+    def Maximum(State, Ply_num, Alpha):
+        if Ply_num == 0:
+            return State.CurrentScore
+
+        for i in range(State.Current.dimY):
+            for j in range(State.Current.dimX):
+                if State.Current.Mat[i][j] == ' ' and (j, i) not in State.children:
+                    State.Make(j, i, False)
+
+        Maximum_Score = -1000
+        i = 0
+        j = 0
+        for k, z in State.children.items():
+            Result = Algo.Minimum(z, Ply_num - 1, Maximum_Score)
+            if Maximum_Score < Result:
+                Maximum_Score = Result
+            if Result > Alpha:
+                return Result
+
+        return Maximum_Score
+
+    def Minimum(State, Ply_num, Beta):  # Alpha-beta pruning function for taking care of Beta values
+        if Ply_num == 0:
+            return State.CurrentScore
+
+        for i in range(State.Current.dimY):
+            for j in range(State.Current.dimX):
+                if State.Current.Mat[i][j] == ' ' and (j, i) not in State.children:
+                    State.Make(j, i, True)
+
+        Minimum_Score = 1000
+        i = 0
+        j = 0
+        for k, z in State.children.items():
+            Result = Algo.Maximum(z, Ply_num - 1, Minimum_Score)
+            if Minimum_Score > Result:
+                Minimum_Score = Result
+            if Result < Beta:
+                return Result
+
+        return Minimum_Score
diff --git a/dotsandboxes/agents/algorithms/ann.py b/dotsandboxes/agents/algorithms/ann.py
new file mode 100644
index 0000000..05ae647
--- /dev/null
+++ b/dotsandboxes/agents/algorithms/ann.py
@@ -0,0 +1,170 @@
+from numpy import *
+from math import sqrt
+from copy import deepcopy
+from time import time
+
+class ANN:
+	
+	"""ANN with one hidden layer, one output and full connections in between consecutive layers.
+	Initial weights are chosen from a normal distribution.
+	Activation function is tanh."""
+	
+	INIT_SIGMA = 0.02
+	REL_STOP_MARGIN = 0.01
+	MAX_ITERATIONS = 1000000
+	ACTIVATION = tanh
+	D_ACTIVATION = lambda x: 1 - tanh(x)**2 # Derivative of tanh
+	VEC_ACTIVATION = vectorize(ACTIVATION)
+	VEC_D_ACTIVATION = vectorize(D_ACTIVATION)
+	STEP_SIZE = 0.1
+	
+	def __init__(self, input_size, hidden_size):
+		
+		#self.input_size = input_size
+		#self.hidden_size = hidden_size
+		self.hidden_weights = random.normal(0, ANN.INIT_SIGMA, (hidden_size, input_size))
+		self.output_weights = random.normal(0, ANN.INIT_SIGMA, hidden_size)
+	
+	def get_weights(self):
+		return self.hidden_weights, self.output_weights
+	
+	def predict(self, input_vector):
+		
+		# Predicts the output for this input vector
+		# input_vector will be normalized
+		
+		input_vector = input_vector/linalg.norm(input_vector)
+		return ANN.ACTIVATION(dot(self.output_weights, ANN.VEC_ACTIVATION(dot(self.hidden_weights, input_vector))))
+	
+	@staticmethod
+	def frob_norm(a, b):
+		
+		# Calculates the total Frobenius norm of both matrices A and B
+		return sqrt(linalg.norm(a)**2 + linalg.norm(b)**2)
+	
+	def train(self, examples):
+		
+		#print("Training")
+		start = time()
+		
+		# examples is a list of (input, output)-tuples
+		# input will be normalized
+		# We stop when the weights have converged within some relative margin
+		
+		for example in examples:
+			example[0] = example[0]/linalg.norm(example[0])
+		
+		iteration = 0
+		while True:
+			
+			
+			# Store old weights to check for convergence later
+			prev_hidden_weights = deepcopy(self.hidden_weights)
+			prev_output_weights = deepcopy(self.output_weights)
+			
+			for k in range(len(examples)):
+				
+				input_vector, output = examples[k]
+				
+				# Calculate outputs
+				hidden_input = dot(self.hidden_weights, input_vector)
+				hidden_output = ANN.VEC_ACTIVATION(hidden_input)
+				final_input = dot(self.output_weights, hidden_output)
+				predicted_output = ANN.ACTIVATION(final_input)
+				
+				#print("Output:", output)
+				#print("Predicted output:", predicted_output)
+				
+				# Used in calculations
+				prediction_error = output - predicted_output
+				output_derivative = ANN.D_ACTIVATION(final_input)
+				
+				# Adjust output weights and calculate requested hidden change
+				requested_hidden_change = prediction_error*output_derivative*self.output_weights
+				self.output_weights = self.output_weights + ANN.STEP_SIZE*prediction_error*hidden_output
+				
+				#print("After adjusting output weights:", ANN.ACTIVATION(dot(self.output_weights, hidden_output)))
+				
+				# Backpropagate requested hidden change to adjust hidden weights
+				self.hidden_weights = self.hidden_weights + ANN.STEP_SIZE*outer(requested_hidden_change*(ANN.VEC_D_ACTIVATION(hidden_input)), input_vector)
+				
+				#print("After adjusting hidden weights:", ANN.ACTIVATION(dot(self.output_weights, ANN.VEC_ACTIVATION(dot(self.hidden_weights, input_vector)))))
+				
+				# Check stop criteria
+				iteration += 1
+				if iteration >= ANN.MAX_ITERATIONS:
+					break
+			
+			# Check stop criteria
+			if iteration >= ANN.MAX_ITERATIONS:
+				break
+			diff = ANN.frob_norm(self.hidden_weights - prev_hidden_weights, self.output_weights - prev_output_weights)
+			base = ANN.frob_norm(self.hidden_weights, self.output_weights)
+			#if base > 0 and diff/base < ANN.REL_STOP_MARGIN:
+			#	break
+		
+		print(time() - start)
+		print("Stopped training after %s iterations."%iteration)
+
+# TESTING
+
+def print_difference(ann1, ann2):
+	
+	# Prints the differences in weights in between two ANN's with identical topology
+	
+	hidden_weights1, output_weights1 = ann1.get_weights()
+	hidden_weights2, output_weights2 = ann2.get_weights()
+	hidden_diff = hidden_weights1 - hidden_weights2
+	output_diff = output_weights1 - output_weights2
+	
+	print(hidden_diff)
+	print(output_diff)
+	print("Frobenius norms:")
+	print("Hidden weights difference:", linalg.norm(hidden_diff))
+	print("Output weights difference:", linalg.norm(output_diff))
+	print("Both:", ANN.frob_norm(hidden_diff, output_diff))
+
+def RMSE(ann, examples):
+	
+	total = 0
+	for input_vector, output in examples:
+		total += (output - ann.predict(input_vector))**2
+	return sqrt(total/len(examples))
+
+def generate_examples(amount, input_size, evaluate):
+	# evaluate is a function mapping an input vector onto a numerical value
+	examples = []
+	inputs = random.normal(0, 100, (amount, input_size))
+	for i in range(amount):
+		input_vector = inputs[i]
+		examples.append([input_vector, evaluate(input_vector)])
+	return examples
+
+def test():
+	
+	# Test the ANN by having it model another ANN with identical topology but unknown weights
+	
+	input_size = 5
+	hidden_size = 3
+	real = ANN(input_size, hidden_size)
+	model = ANN(input_size, hidden_size)
+	
+	# Generate training data
+	training_data = generate_examples(10000, input_size, real.predict)
+	validation_data = generate_examples(10000, input_size, real.predict)
+	
+	# Print initial difference, train, then print new difference
+	print("Initial difference:")
+	print_difference(real, model)
+	print("Initial RMSE (on training data):", RMSE(model, training_data))
+	print("Initial RMSE (on validation data):", RMSE(model, validation_data))
+	model.train(training_data)
+	print("After training:")
+	print_difference(real, model)
+	print("After training RMSE (on training data):", RMSE(model, training_data))
+	print("After training RMSE (on validation data):", RMSE(model, validation_data))
+
+if __name__ == "__main__":
+	test()
+
+