edited syllabus, checked deps

pyproject.toml

@@ -1,3 +1,6 @@
+[build-system]
+requires = ["pip>=23.0", "setuptools>=61.0", "wheel"]
+
 [project]
 name = "nmmo2-baselines"
 version = "0.1.0"
@@ -14,10 +17,11 @@ classifiers = [
 dependencies = [
     "accelerate==0.27.2",
     "nmmo@git+https://github.com/kywch/nmmo-environment",  # WIP nmmo 2.1
-    "polars",
-    "pufferlib[nmmo]==0.7.3",
-    "psutil",
-    "torch",
+    "polars==0.20.21",
+    "pufferlib[nmmo]>=0.7.3",
+    "psutil==5.9.8",
+    "syllabus-rl@git+https://github.com/kywch/Syllabus@nmmo",  # To replace with pip later
+    "torch==2.1.0",  # match puffertank
     "transformers==4.37.2",
     "wandb",
 ]

reinforcement_learning/environment.py

@@ -6,12 +6,9 @@
 import pufferlib
 import pufferlib.emulation
 from pettingzoo.utils.wrappers.base_parallel import BaseParallelWrapper
-from syllabus.core import PettingZooMultiProcessingSyncWrapper
-from syllabus_task_wrapper import NMMOTaskWrapper
+from syllabus.core import PettingZooMultiProcessingSyncWrapper as SyllabusSyncWrapper
 
-
-def alt_combat_damage_formula(offense, defense, multiplier, minimum_proportion):
-    return int(max(multiplier * offense - defense, offense * minimum_proportion))
+from syllabus_wrapper import SyllabusTaskWrapper
 
 
 class Config(
@@ -27,6 +24,7 @@ class Config(
     nc.Exchange,
 ):
     """Configuration for Neural MMO."""
+
     def __init__(self, env_args: Namespace):
         super().__init__()
 
@@ -51,21 +49,25 @@ def __init__(self, env_args: Namespace):
         self.set("CURRICULUM_FILE_PATH", env_args.curriculum_file_path)
 
 
-def make_env_creator(reward_wrapper_cls: BaseParallelWrapper, task_wrapper=False, curriculum=None):
+def make_env_creator(
+    reward_wrapper_cls: BaseParallelWrapper, syllabus_wrapper=False, syllabus=None
+):
     def env_creator(*args, **kwargs):
         """Create an environment."""
         env = nmmo.Env(Config(kwargs["env"]))  # args.env is provided as kwargs
         env = reward_wrapper_cls(env, **kwargs["reward_wrapper"])
 
         # Add Syllabus task wrapper
-        if task_wrapper or curriculum is not None:
-            env = NMMOTaskWrapper(env)
+        if syllabus_wrapper or syllabus is not None:
+            env = SyllabusTaskWrapper(env)
 
-        # Use curriculum if provided
-        if curriculum is not None:
-            # Add Syllabus Sync Wrapper
-            env = PettingZooMultiProcessingSyncWrapper(
-                env, curriculum.get_components(), update_on_step=False, task_space=env.task_space,
+        # Use syllabus curriculum if provided
+        if syllabus is not None:
+            env = SyllabusSyncWrapper(
+                env,
+                syllabus.get_components(),
+                update_on_step=False,
+                task_space=env.task_space,
             )
 
         env = pufferlib.emulation.PettingZooPufferEnv(env)