Spaces:
Sleeping
Sleeping
File size: 9,619 Bytes
4a38f8e 4e2d608 4a38f8e 4e2d608 4a38f8e 4e2d608 4a38f8e 4e2d608 4a38f8e 4e2d608 4a38f8e 4e2d608 e0b7d34 4e2d608 e0b7d34 4e2d608 e0b7d34 4a38f8e e0b7d34 4e2d608 e0b7d34 4e2d608 e0b7d34 4a38f8e e0b7d34 4e2d608 e0b7d34 4e2d608 e0b7d34 4e2d608 e0b7d34 4e2d608 e0b7d34 4e2d608 e0b7d34 4a38f8e e0b7d34 4a38f8e 4e2d608 46da75d 4e2d608 46da75d 4a38f8e 4e2d608 4a38f8e 4e2d608 4a38f8e 99a9aba 4a38f8e d19f083 4a38f8e 4e2d608 4a38f8e 4e2d608 4a38f8e 4e2d608 4a38f8e 4e2d608 4a38f8e 46da75d 4e2d608 4a38f8e 4e2d608 46da75d 2e3b252 46da75d 647ef48 4e2d608 46da75d 4e2d608 4a38f8e 4e2d608 4a38f8e a820e84 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 | from fastapi import FastAPI, HTTPException
from pydantic import BaseModel
from typing import Dict, List, Optional, Any
import ast
import uuid
import threading
from tasks import TASKS
import uvicorn
# --- Models ---
class CodeMigrationAction(BaseModel):
func_id: str
rewritten_code: str
class CodeMigrationObservation(BaseModel):
pending_functions: List[Dict[str, Any]]
completed_functions: List[str]
target_library: str
last_error: Optional[str] = None
class Reward(BaseModel):
value: float
# --- Environment ---
class CodeMigrationEnv:
def __init__(self, task_config: Dict):
self.config = task_config
self.pending = {f["func_id"]: f for f in task_config["functions"]}
self.completed = []
self.last_error = None
self.cumulative_reward = 0.0
def _uses_old_library(self, code_str: str) -> bool:
source_prefix = self.config["source_lib"].split('.')[0]
try:
tree = ast.parse(code_str)
for node in ast.walk(tree):
if isinstance(node, (ast.Import, ast.ImportFrom)):
names = [n.name for n in node.names]
if any(name.startswith(source_prefix) for name in names): return True
if isinstance(node, ast.ImportFrom) and node.module and node.module.startswith(source_prefix): return True
return source_prefix in code_str
except: return True
def calculate_complex_reward(self, old_code: str, new_code: str) -> float:
import ast
target_lib = self.config["target_lib"]
source_lib = self.config["source_lib"]
reward = 0.0
# --- Weights (tunable) ---
W = {
"migration": 0.25,
"target_usage": 0.15,
"semantic_similarity": 0.20,
"signature": 0.10,
"async_correctness": 0.10,
"error_handling": 0.05,
"code_quality": 0.10,
"penalty_complexity": -0.10
}
try:
old_tree = ast.parse(old_code)
new_tree = ast.parse(new_code)
# ----------------------------
# 1. Migration correctness
# ----------------------------
if not self._uses_old_library(new_code):
reward += W["migration"]
else:
self.last_error = f"Still uses {source_lib}"
return -0.3
# ----------------------------
# 2. Target library usage
# ----------------------------
target_prefix = target_lib.split('.')[0]
uses_target = any(
isinstance(n, (ast.Import, ast.ImportFrom)) and
(
any(name.name.startswith(target_prefix) for name in getattr(n, "names", [])) or
(hasattr(n, "module") and n.module and n.module.startswith(target_prefix))
)
for n in ast.walk(new_tree)
)
if uses_target:
reward += W["target_usage"]
else:
reward -= 0.05 # weak penalty
# ----------------------------
# 3. Semantic similarity (AST overlap heuristic)
# ----------------------------
def get_node_types(tree):
return set(type(n).__name__ for n in ast.walk(tree))
old_nodes = get_node_types(old_tree)
new_nodes = get_node_types(new_tree)
overlap = len(old_nodes & new_nodes) / max(1, len(old_nodes))
reward += W["semantic_similarity"] * overlap
# ----------------------------
# 4. Function signature preservation
# ----------------------------
def get_func_sig(tree):
for node in ast.walk(tree):
if isinstance(node, ast.FunctionDef):
return len(node.args.args)
return 0
if get_func_sig(old_tree) == get_func_sig(new_tree):
reward += W["signature"]
else:
reward -= 0.05
# ----------------------------
# 5. Async correctness (graded, not binary)
# ----------------------------
is_async_target = "async" in target_lib.lower()
async_count = sum(isinstance(n, ast.Await) for n in ast.walk(new_tree))
async_def = any(isinstance(n, ast.AsyncFunctionDef) for n in ast.walk(new_tree))
if is_async_target:
if async_def:
reward += W["async_correctness"] * 0.6
if async_count > 0:
reward += W["async_correctness"] * 0.4
else:
if async_def or async_count > 0:
reward -= 0.05 # unnecessary async usage
# ----------------------------
# 6. Error handling improvement
# ----------------------------
has_try = any(isinstance(n, ast.Try) for n in ast.walk(new_tree))
if has_try:
reward += W["error_handling"]
# ----------------------------
# 7. Code quality / idioms
# ----------------------------
quality_score = 0.0
# context managers
if any(isinstance(n, (ast.With, ast.AsyncWith)) for n in ast.walk(new_tree)):
quality_score += 0.4
# list/dict comprehensions
if any(isinstance(n, (ast.ListComp, ast.DictComp)) for n in ast.walk(new_tree)):
quality_score += 0.3
# avoids excessive nesting
max_depth = 0
def get_depth(node, depth=0):
nonlocal max_depth
max_depth = max(max_depth, depth)
for child in ast.iter_child_nodes(node):
get_depth(child, depth + 1)
get_depth(new_tree)
if max_depth < 10:
quality_score += 0.3
reward += W["code_quality"] * quality_score
# ----------------------------
# 8. Complexity penalty
# ----------------------------
def count_nodes(tree):
return sum(1 for _ in ast.walk(tree))
old_size = count_nodes(old_tree)
new_size = count_nodes(new_tree)
if new_size > old_size * 1.5:
reward += W["penalty_complexity"]
# ----------------------------
# Final normalization
# ----------------------------
self.last_error = None
return round(max(-1.0, min(1.0, reward)), 3)
except SyntaxError as e:
self.last_error = f"Syntax Error: {e}"
return -0.6
def step(self, action: CodeMigrationAction):
if action.func_id not in self.pending:
return self.state(), Reward(value=-0.1), False, {"error": "Invalid ID"}
old_code = self.pending[action.func_id]["code"]
reward_value = self.calculate_complex_reward(old_code, action.rewritten_code)
# Move to completed only if functional success was achieved (reward > 0)
if reward_value > 0:
reward_value = max(0.10, min(0.90, reward_value))
self.completed.append(action.func_id)
del self.pending[action.func_id]
else:
# Penalties stay negative, which is fine
reward_value = min(-0.01, reward_value)
done = len(self.pending) == 0
return self.state(), Reward(value=reward_value), done, {}
def state(self) -> CodeMigrationObservation:
return CodeMigrationObservation(
pending_functions=list(self.pending.values()),
completed_functions=self.completed,
target_library=self.config["target_lib"],
last_error=self.last_error
)
# --- FastAPI App ---
app = FastAPI()
sessions: Dict[str, Dict] = {}
_lock = threading.Lock()
@app.get("/")
def ping():
return {"status": "ok", "message": "space is running"}
@app.post("/reset")
def reset(task_id: str = "task_easy"):
if task_id not in TASKS: raise HTTPException(404)
env = CodeMigrationEnv(TASKS[task_id])
sid = str(uuid.uuid4())
with _lock: sessions[sid] = {"env": env, "rewards": []}
return {"session_id": sid, "observation": env.state().model_dump()}
@app.post("/step")
def step(session_id: str, action: CodeMigrationAction):
with _lock:
if session_id not in sessions: raise HTTPException(404)
env = sessions[session_id]["env"]
obs, reward, done, info = env.step(action)
with _lock: sessions[session_id]["rewards"].append(reward.value)
return {"observation": obs.model_dump(), "reward": reward.model_dump(), "done": done, "info": info}
@app.get("/grader")
def grader(session_id: str):
with _lock:
session = sessions.get(session_id)
if not session: raise HTTPException(404)
total = sum(session["rewards"])
n_funcs = len(TASKS[session["env"].config["id"]]["functions"])
# Calculate raw average
raw_average = total / n_funcs if n_funcs > 0 else 0.01
score = max(0.10, min(0.90, raw_average))
return {
"score": float(f"{score:.3f}"), # Force 3 decimal places as a float
"success": len(session["env"].pending) == 0
}
def main():
import os
port = int(os.environ.get("PORT", 7860))
uvicorn.run(app, host="0.0.0.0", port=port)
if __name__ == "__main__":
main() |