Revert "Merge pull request #25 from paperswithbacktest/codex/improve-llm-to-find-better-trading-strategy"

This reverts commit 57bda36, reversing changes made to d953241.

Author: edarchimbaud

README.md

@@ -116,8 +116,8 @@ python scripts/run_example.py --experiment my_exp
 | Data | Zero‑setup loader for any Papers‑With‑Backtest dataset (`pwb_toolbox`) + caching to Feather |
 | Strategies | Seed templates with **EVOLVE‑BLOCK** markers that the LLM mutates |
 | Evaluator | Deterministic Backtrader walk‑forward, JSON KPIs (Sharpe, CAGR, Calmar, DD) |
-| LLM Engine | OpenAI o3 structured‑output chat → JSON diff/patch system; prompts push each child to beat its parent |
-| Evolution | Async controller stores a child only when it improves the chosen metric; SQLite hall‑of‑fame, optional MAP‑Elites niches |
+| LLM Engine | OpenAI o3 structured‑output chat → JSON diff/patch system  |
+| Evolution | Async controller, SQLite hall‑of‑fame, optional MAP‑Elites niches  |
 | Dashboard | (optional) Streamlit live view of metrics & equity curves |
 
 ---

alphaevolve/evolution/controller.py

@@ -134,26 +134,14 @@ async def _spawn(self, parent_id: str | None, *, prompt: PromptGenome | None = N
  logger.error(f"Evaluation failed: {e}")
  return
 
- # 5) Persist if improved
- parent_metrics = parent.get("metrics") or {}
- parent_val = parent_metrics.get(self.metric)
- child_val = kpis.get(self.metric)
-
- if child_val is None or parent_val is None or child_val > parent_val:
- self.store.insert(
- child_code,
- kpis,
- parent_id=parent["id"],
- island=parent.get("island", 0),
- )
- logger.info("Child stored (%s %.2f)", self.metric, kpis.get(self.metric, 0))
- else:
- logger.info(
- "Discarded child (%s %.2f <= %.2f)",
- self.metric,
- child_val,
- parent_val,
- )
+ # 5) Persist
+ self.store.insert(
+ child_code,
+ kpis,
+ parent_id=parent["id"],
+ island=parent.get("island", 0),
+ )
+ logger.info("Child stored (%s %.2f)", self.metric, kpis.get(self.metric, 0))
 
  # ------------------------------------------------------------------
  # public API

alphaevolve/llm_engine/prompts.py

@@ -25,8 +25,7 @@
 
 SYSTEM_MSG = """\
 You are Alpha-Trader Evolution-Engine. You mutate algorithmic-trading
-strategies written for Backtrader. Each new iteration should attempt
-to outperform its parent based on risk-adjusted metrics. All editable regions are delimited
+strategies written for Backtrader. All editable regions are delimited
 like this:
 
  # === EVOLVE-BLOCK: <block_name> =================================

tests/test_controller.py

@@ -211,27 +211,3 @@ def test_controller_respects_population_limit(tmp_path):
  assert store._count() <= 2
  finally:
  _cleanup(installed)
-
-
-def test_controller_discards_unimproved_child(tmp_path):
- diff = '{"blocks": {"logic": "b = 2"}}'
- metrics = {"sharpe": 1.0}
- ctrl, store, installed = _setup_controller(tmp_path, diff, metrics)
- try:
- # first spawn inserts child with sharpe 1.0
- asyncio.run(_run_spawn(ctrl))
- child_id = store.top_k(k=1)[0]["id"]
-
- # make evaluator return worse metrics
- async def evaluate(_code, *, symbols=None):
- return {"sharpe": 0.5, "calmar": 0.0, "cagr": 0.0}
-
- sys.modules["alphaevolve.evaluator.backtest"].evaluate = evaluate
-
- # spawn from the existing child
- asyncio.run(ctrl._spawn(child_id))
-
- # count should remain 2 since new child discarded
- assert store._count() == 2
- finally:
- _cleanup(installed)