Setup local

.gitignore

@@ -62,6 +62,7 @@ environment/src/skills/ground_truth_skills/*
 /eval/open/mcts/runs/
 /eval/open/independent_runs/summary_cache/
 /eval/open/beam/summary_cache/
+/summary_cache/
 /venv_py311/
 /eval/open/plots/
 *.mp4

env/src/instance.py

@@ -322,14 +322,16 @@ def eval_with_error(self, expr, timeout=60):
         def handler(signum, frame):
             raise TimeoutError()
 
-        signal.signal(signal.SIGALRM, handler)
-        signal.alarm(timeout)
+        # signal.signal(signal.SIGALRM, handler)
+        # signal.alarm(timeout)
+        alarm = threading.Timer(timeout, handler)
+        alarm.start()
 
         try:
             return self.namespace.eval_with_timeout(expr)
         finally:
-            signal.alarm(0)
-
+            # signal.alarm(0)
+            alarm.cancel()
 
     def eval(self, expr, timeout=60):
         "Evaluate several lines of input, returning the result of the last line with a timeout"

eval/open/beam/run_parallel.py

@@ -16,6 +16,7 @@
 from instance import FactorioInstance
 from agents.utils.formatters.recursive_report_formatter import RecursiveReportFormatter
 from models.game_state import GameState
+from eval.open.db_client import SQLliteDBClient
 
 os.environ.update({"FORCE_COLOR": "1", "TERM": "xterm-256color"})
 load_dotenv()
@@ -73,23 +74,30 @@ async def get_version_to_use(resume_version: int = None) -> int:
 
 
 async def create_db_client() -> DBClient:
-    """Create and return a new database client."""
-    try:
-        max_connections = 5  # Per process
-        min_connections = 2
-        return DBClient(
-            max_conversation_length=40,
-            min_connections=min_connections,
-            max_connections=max_connections,
-            host=os.getenv("SKILLS_DB_HOST"),
-            port=os.getenv("SKILLS_DB_PORT"),
-            dbname=os.getenv("SKILLS_DB_NAME"),
-            user=os.getenv("SKILLS_DB_USER"),
-            password=os.getenv("SKILLS_DB_PASSWORD")
-        )
-    except Exception as e:
-        print(f"\033[91mError connecting to the database: {e}\033[91m")
-        raise
+    return SQLliteDBClient(
+        max_conversation_length=40,
+        min_connections=2,
+        max_connections=5,
+        database_file=os.getenv("SQLITE_DB_FILE"),
+    )
+
+    # """Create and return a new database client."""
+    # try:
+    #     max_connections = 5  # Per process
+    #     min_connections = 2
+    #     return DBClient(
+    #         max_conversation_length=40,
+    #         min_connections=min_connections,
+    #         max_connections=max_connections,
+    #         host=os.getenv("SKILLS_DB_HOST"),
+    #         port=os.getenv("SKILLS_DB_PORT"),
+    #         dbname=os.getenv("SKILLS_DB_NAME"),
+    #         user=os.getenv("SKILLS_DB_USER"),
+    #         password=os.getenv("SKILLS_DB_PASSWORD")
+    #     )
+    # except Exception as e:
+    #     print(f"\033[91mError connecting to the database: {e}\033[91m")
+    #     raise
 
 
 async def run_model_search(model: str, instance_start: int, version: int, resume_version: int = None):

eval/open/independent_runs/run_config_example_lab_play.json

@@ -1,4 +1,4 @@
 [
-    {"task": "iron_ore_throughput_16.json",
-    "model": "gpt-4o"}
+    {"task": "electronic_circuit_throughput_16.json",
+    "model": "open-router-anthropic/claude-3.5-sonnet"}
     ]

eval/open/independent_runs/trajectory_runner.py

@@ -10,7 +10,7 @@
 from agents import CompletionResult, CompletionReason
 from agents.agent_abc import AgentABC
 from agents.basic_agent import BasicAgent
-from eval.open.db_client import PostgresDBClient, SQLliteDBClient
+from eval.open.db_client import PostgresDBClient, SQLliteDBClient, DBClient
 from eval.open.independent_runs.simple_evaluator import SimpleFactorioEvaluator
 from models.conversation import Conversation
 from models.message import Message
@@ -244,19 +244,26 @@ def create_factorio_instance(instance_id: int) -> FactorioInstance:
     return instance
 
 
-async def create_db_client() -> PostgresDBClient:
-    """Create database client with connection pool"""
-    return PostgresDBClient(
+async def create_db_client() -> DBClient:
+    return SQLliteDBClient(
         max_conversation_length=40,
         min_connections=2,
         max_connections=5,
-        host=os.getenv("SKILLS_DB_HOST"),
-        port=os.getenv("SKILLS_DB_PORT"),
-        dbname=os.getenv("SKILLS_DB_NAME"),
-        user=os.getenv("SKILLS_DB_USER"),
-        password=os.getenv("SKILLS_DB_PASSWORD")
+        database_file=os.getenv("SQLITE_DB_FILE"),
     )
 
+    # """Create database client with connection pool"""
+    # return PostgresDBClient(
+    #     max_conversation_length=40,
+    #     min_connections=2,
+    #     max_connections=5,
+    #     host=os.getenv("SKILLS_DB_HOST"),
+    #     port=os.getenv("SKILLS_DB_PORT"),
+    #     dbname=os.getenv("SKILLS_DB_NAME"),
+    #     user=os.getenv("SKILLS_DB_USER"),
+    #     password=os.getenv("SKILLS_DB_PASSWORD")
+    # )
+
 
 async def run_trajectory(process_id: int, config: EvalConfig):
     """Entry point for running a single trajectory"""

pyproject.toml

@@ -0,0 +1,28 @@
+[project]
+name = "factorio-learning-environment"
+version = "0.1.0"
+description = "Factorio Learning Environment"
+readme = "README.md"
+requires-python = ">=3.10"
+dependencies = [
+    "dotenv>=0.9.9",
+    "pydantic>=2.10.6",
+    "tenacity>=9.0.0",
+    "anthropic>=0.49.0",
+    "openai>=1.66.3",
+    "psycopg2>=2.9.10",
+    "numpy>=2.2.3",
+    "lupa>=2.4",
+    "slpp>=1.2.3",
+    "factorio-rcon-py>=2.1.3",
+    "construct>=2.10.70",
+    "pillow>=11.1.0",
+    "scikit-image>=0.25.2",
+]
+
+[build-system]
+requires = ["hatchling"]
+build-backend = "hatchling.build"
+
+[tool.hatch.build.targets.wheel]
+packages = ["agents", "env/src/models"]