Version 1.2.0 (#8)

montecarlo/montecarlo.py

@@ -25,7 +25,7 @@ def simulate(self, expansion_count = 1):
 			current_node = self.root_node
 
 			while current_node.expanded:
-				current_node = current_node.get_preferred_child()
+				current_node = current_node.get_preferred_child(self.root_node)
 
 			self.expand(current_node)
 

montecarlo/node.py

@@ -11,6 +11,7 @@ def __init__(self, state):
 		self.parent = None
 		self.children = []
 		self.expanded = False
+		self.player_number = None
 
 	def update_win_value(self, value):
 		self.win_value += value
@@ -30,12 +31,12 @@ def add_children(self, children):
 		for child in children:
 			self.add_child(child)
 
-	def get_preferred_child(self):
+	def get_preferred_child(self, root_node):
 		best_children = []
 		best_score = float('-inf')
 
 		for child in self.children:
-			score = child.get_score()
+			score = child.get_score(root_node)
 
 			if score > best_score:
 				best_score = score
@@ -45,10 +46,12 @@ def get_preferred_child(self):
 
 		return random.choice(best_children)
 
-	def get_score(self):
+	def get_score(self, root_node):
 		discovery_constant = 0.35
 		discovery_operand = discovery_constant * (self.policy_value or 1) * sqrt(log(self.parent.visits) / (self.visits or 1))
-		win_operand = self.win_value / (self.visits or 1)
+
+		win_multiplier = 1 if self.parent.player_number == root_node.player_number else -1
+		win_operand = win_multiplier * self.win_value / (self.visits or 1)
 
 		self.score = win_operand + discovery_operand
 

readme.md

@@ -1,6 +1,6 @@
 A Python3 library that you can use to run a Monte Carlo tree search, either traditionally with drilling down to end game states or with expert policies as you might provide from a neural network.
 
-- **Version:** 1.1.2
+- **Version:** 1.2.0
 
 [![Build Status](https://travis-ci.org/ImparaAI/monte-carlo-tree-search.png?branch=master)](https://travis-ci.org/ImparaAI/monte-carlo-tree-search)
 
@@ -90,4 +90,13 @@ After you've chosen a new root node, you can override it on the `montecarlo` ins
 
 ```python
 montecarlo.root_node = montecarlo.make_choice()
-```
+```
+
+## Turn based environments
+
+If you are modeling a turn based environment (E.g. a two player board game), set the player_number on each node in order for the selection process to invert child win values.
+
+```python
+node = Node(state)
+node.player_number = 1
+```

setup.py

@@ -5,7 +5,7 @@
 
 setuptools.setup(
     name="imparaai-montecarlo",
-    version="1.1.2",
+    version="1.2.0",
     license='MIT',
     author="ImparaAI",
     author_email="[email protected]",