Added stage completion handling

celeste/celeste_ai/celeste.py

@@ -50,7 +50,6 @@ class Celeste:
 	action_space = [
 		"left",		# move left		0
 		"right",	# move right	1
-		#"jump",	# jump
 		"jump-l",	# jump left		2
 		"jump-r",	# jump right	3
 
@@ -86,6 +85,13 @@ class Celeste:
 		]
 	]
 
+
+	# Maps room_x, room_y coordinates to stage number.
+	stage_map = [
+		[0, 1, 2, 3, 4]
+	]
+
+
 	def __init__(
 			self,
 			pico_path,
@@ -194,9 +200,7 @@ class Celeste:
 	def state(self):
 		try:
 			stage = (
-				[
-					[0, 1, 2, 3, 4]
-				]
+				Celeste.stage_map
 				[int(self._internal_state["ry"])]
 				[int(self._internal_state["rx"])]
 			)

celeste/celeste_ai/train.py

@@ -341,10 +341,23 @@ def on_state_after(celeste, before_out):
 		dtype = torch.long
 	)
 
+	finished_stage = False
+	# No reward if dead
 	if next_state.deaths != 0:
 		pt_next_state = None
 		reward = 0
 
+	# Reward for finishing stage
+	elif next_state.stage >= 1:
+		finished_stage = True
+		reward = next_state.next_point - state.next_point
+		reward += 1
+
+		# Add to point counter
+		for i in range(state.next_point, state.next_point + reward):
+			point_counter[i] += 1
+
+	# Regular reward
 	else:
 		pt_next_state = torch.tensor(
 			[getattr(next_state, x) for x in Celeste.state_number_map],
@@ -401,7 +414,7 @@ def on_state_after(celeste, before_out):
 
 	# Move on to the next episode once we reach
 	# a terminal state.
-	if (next_state.deaths != 0):
+	if (next_state.deaths != 0 or finished_stage):
 		s = celeste.state
 		with model_train_log.open("a") as f:
 			f.write(json.dumps({