claude-turing 2.3.0 → 2.4.0

package/templates/scripts/generate_brief.py

@@ -309,6 +309,23 @@ def load_regression_checks(regress_dir: str = "experiments/regressions") -> list
     return reports
 
 
+def load_ensemble_results(ensemble_dir: str = "experiments/ensembles") -> list[dict]:
+    """Load ensemble result reports from YAML files."""
+    path = Path(ensemble_dir)
+    if not path.exists():
+        return []
+    reports = []
+    for f in sorted(path.glob("ensemble-*.yaml")):
+        try:
+            with open(f) as fh:
+                report = yaml.safe_load(fh)
+                if report and isinstance(report, dict):
+                    reports.append(report)
+        except (yaml.YAMLError, OSError):
+            continue
+    return reports
+
+
 def format_brief(
     campaign: dict,
     best: dict | None,
@@ -327,6 +344,7 @@ def format_brief(
     profiles: list[dict] | None = None,
     queue_summary: dict | None = None,
     regression_checks: list[dict] | None = None,
+    ensemble_results: list[dict] | None = None,
 ) -> str:
     """Format the research briefing as markdown."""
     direction = "lower" if lower_is_better else "higher"
@@ -546,6 +564,21 @@ def format_brief(
         if auto_hyps:
             lines.append(f"\n*{auto_hyps} auto-generated hypotheses from failure analysis.*")
 
+    # Ensemble results
+    if ensemble_results:
+        lines.extend(["", "## Ensembles", ""])
+        for ens in ensemble_results:
+            best_method = ens.get("best_method", "?")
+            improvement = ens.get("improvement", 0)
+            n_models = ens.get("n_candidates", 0)
+            if best_method != "best_single" and improvement > 0:
+                lines.append(
+                    f"- **{best_method}** ({n_models} models): "
+                    f"{metric} improvement {improvement:+.4f} over best single"
+                )
+            else:
+                lines.append(f"- {n_models}-model ensemble: no improvement over best single")
+
     # Regression check history (stability)
     if regression_checks:
         lines.extend(["", "## Stability", ""])
@@ -636,6 +669,7 @@ def generate_brief(
     profiles = load_profiles()
     queue_summary = load_queue_summary()
     regression_checks = load_regression_checks()
+    ensemble_results = load_ensemble_results()
 
     return format_brief(
         campaign, best, trajectory, model_types, hypotheses,
@@ -648,6 +682,7 @@ def generate_brief(
         profiles=profiles if profiles else None,
         queue_summary=queue_summary,
         regression_checks=regression_checks if regression_checks else None,
+        ensemble_results=ensemble_results if ensemble_results else None,
     )
 
 

package/templates/scripts/pipeline_manager.py

@@ -0,0 +1,457 @@
+#!/usr/bin/env python3
+"""Pipeline composition manager for the autoresearch pipeline.
+
+Decomposes ML pipelines into swappable stages (preprocess, features,
+model, postprocess). Each stage can be independently varied, cached,
+and reused across experiments.
+
+Usage:
+    python scripts/pipeline_manager.py show
+    python scripts/pipeline_manager.py swap model --from exp-031
+    python scripts/pipeline_manager.py cache
+    python scripts/pipeline_manager.py run
+"""
+
+from __future__ import annotations
+
+import argparse
+import hashlib
+import json
+import sys
+from datetime import datetime, timezone
+from pathlib import Path
+
+import yaml
+
+from scripts.turing_io import load_config, load_experiments
+
+DEFAULT_LOG_PATH = "experiments/log.jsonl"
+DEFAULT_CACHE_DIR = "experiments/cache"
+DEFAULT_STAGES = ["preprocess", "features", "model", "postprocess"]
+
+
+# --- Stage Definition ---
+
+
+def define_stages(config: dict) -> list[dict]:
+    """Extract pipeline stage definitions from config.
+
+    Args:
+        config: Project config dict.
+
+    Returns:
+        List of stage dicts with name, type, config_keys, and description.
+    """
+    pipeline_cfg = config.get("pipeline", {})
+    stage_names = pipeline_cfg.get("stages", DEFAULT_STAGES)
+
+    stages = []
+    for name in stage_names:
+        stage = {
+            "name": name,
+            "config": pipeline_cfg.get(name, {}),
+            "description": _describe_stage(name, config),
+        }
+        stage["hash"] = compute_stage_hash(stage)
+        stages.append(stage)
+
+    return stages
+
+
+def _describe_stage(name: str, config: dict) -> str:
+    """Generate a human-readable description of a pipeline stage."""
+    if name == "preprocess":
+        model_cfg = config.get("model", {})
+        steps = []
+        if model_cfg.get("hyperparams", {}).get("handle_missing"):
+            steps.append("handle_missing")
+        if model_cfg.get("hyperparams", {}).get("scaler"):
+            steps.append(model_cfg["hyperparams"]["scaler"])
+        return ", ".join(steps) if steps else "default preprocessing"
+
+    elif name == "features":
+        model_cfg = config.get("model", {})
+        features = model_cfg.get("features", [])
+        if features:
+            return ", ".join(features[:5])
+        return "raw features"
+
+    elif name == "model":
+        model_cfg = config.get("model", {})
+        model_type = model_cfg.get("type", "unknown")
+        hyperparams = model_cfg.get("hyperparams", {})
+        key_params = []
+        for k in ("max_depth", "n_estimators", "learning_rate", "hidden_size"):
+            if k in hyperparams:
+                key_params.append(f"{k}={hyperparams[k]}")
+        return f"{model_type}({', '.join(key_params)})" if key_params else model_type
+
+    elif name == "postprocess":
+        return config.get("postprocess", {}).get("type", "none")
+
+    return "unknown"
+
+
+def compute_stage_hash(stage: dict) -> str:
+    """Compute a content hash for a stage configuration.
+
+    Used to detect when a stage hasn't changed and skip re-computation.
+    """
+    content = json.dumps(stage.get("config", {}), sort_keys=True)
+    content += stage.get("description", "")
+    return hashlib.md5(content.encode()).hexdigest()[:8]
+
+
+# --- Stage Operations ---
+
+
+def extract_stage_from_experiment(
+    experiment: dict,
+    stage_name: str,
+) -> dict:
+    """Extract a specific stage's configuration from an experiment.
+
+    Args:
+        experiment: Experiment dict from log.
+        stage_name: Name of the stage to extract.
+
+    Returns:
+        Stage config dict.
+    """
+    config = experiment.get("config", {})
+    hyperparams = config.get("hyperparams", {})
+
+    if stage_name == "model":
+        return {
+            "name": "model",
+            "config": {
+                "type": config.get("model_type", "unknown"),
+                "hyperparams": hyperparams,
+            },
+            "source_experiment": experiment.get("experiment_id"),
+        }
+
+    elif stage_name == "preprocess":
+        preprocess_keys = {"scaler", "handle_missing", "imputer", "normalize"}
+        preprocess_cfg = {k: v for k, v in hyperparams.items() if k in preprocess_keys}
+        return {
+            "name": "preprocess",
+            "config": preprocess_cfg,
+            "source_experiment": experiment.get("experiment_id"),
+        }
+
+    elif stage_name == "features":
+        feature_keys = {"features", "feature_engineering", "polynomial", "interactions"}
+        feature_cfg = {k: v for k, v in hyperparams.items() if k in feature_keys}
+        return {
+            "name": "features",
+            "config": feature_cfg,
+            "source_experiment": experiment.get("experiment_id"),
+        }
+
+    elif stage_name == "postprocess":
+        post_keys = {"calibration", "threshold", "postprocess"}
+        post_cfg = {k: v for k, v in hyperparams.items() if k in post_keys}
+        return {
+            "name": "postprocess",
+            "config": post_cfg,
+            "source_experiment": experiment.get("experiment_id"),
+        }
+
+    return {"name": stage_name, "config": {}, "source_experiment": experiment.get("experiment_id")}
+
+
+def swap_stage(
+    current_stages: list[dict],
+    stage_name: str,
+    new_stage: dict,
+) -> list[dict]:
+    """Replace a stage in the pipeline with a new one.
+
+    Args:
+        current_stages: Current pipeline stages.
+        stage_name: Name of stage to replace.
+        new_stage: New stage configuration.
+
+    Returns:
+        Updated pipeline stages.
+    """
+    result = []
+    for stage in current_stages:
+        if stage["name"] == stage_name:
+            new_stage["hash"] = compute_stage_hash(new_stage)
+            result.append(new_stage)
+        else:
+            result.append(stage)
+    return result
+
+
+# --- Cache Management ---
+
+
+def get_cache_path(stage: dict, cache_dir: str = DEFAULT_CACHE_DIR) -> Path:
+    """Get the cache file path for a stage."""
+    stage_hash = stage.get("hash", compute_stage_hash(stage))
+    return Path(cache_dir) / f"{stage['name']}-{stage_hash}"
+
+
+def check_cache(stage: dict, cache_dir: str = DEFAULT_CACHE_DIR) -> bool:
+    """Check if a stage's output is cached."""
+    cache_path = get_cache_path(stage, cache_dir)
+    return cache_path.exists()
+
+
+def get_cache_stats(cache_dir: str = DEFAULT_CACHE_DIR) -> dict:
+    """Get cache directory statistics."""
+    path = Path(cache_dir)
+    if not path.exists():
+        return {"total_files": 0, "total_size_bytes": 0, "stages": {}}
+
+    total_size = 0
+    total_files = 0
+    stages = {}
+
+    for entry in path.iterdir():
+        if entry.is_dir():
+            size = sum(f.stat().st_size for f in entry.rglob("*") if f.is_file())
+            n_files = sum(1 for f in entry.rglob("*") if f.is_file())
+            stage_name = entry.name.rsplit("-", 1)[0] if "-" in entry.name else entry.name
+            stages[entry.name] = {"size_bytes": size, "n_files": n_files}
+            total_size += size
+            total_files += n_files
+        elif entry.is_file():
+            total_size += entry.stat().st_size
+            total_files += 1
+
+    return {
+        "total_files": total_files,
+        "total_size_bytes": total_size,
+        "total_size_mb": round(total_size / (1024 * 1024), 2),
+        "stages": stages,
+    }
+
+
+# --- Pipeline Composition ---
+
+
+def compose_pipeline(
+    action: str,
+    stage_name: str | None = None,
+    from_exp: str | None = None,
+    config_path: str = "config.yaml",
+    log_path: str = DEFAULT_LOG_PATH,
+    cache_dir: str = DEFAULT_CACHE_DIR,
+) -> dict:
+    """Execute a pipeline composition action.
+
+    Args:
+        action: One of show, swap, cache, run.
+        stage_name: Stage to operate on (for swap).
+        from_exp: Experiment ID to take stage from (for swap).
+        config_path: Path to config.yaml.
+        log_path: Path to experiment log.
+        cache_dir: Cache directory path.
+
+    Returns:
+        Action result dict.
+    """
+    config = load_config(config_path)
+
+    if action == "show":
+        stages = define_stages(config)
+        cache_status = {}
+        for stage in stages:
+            cache_status[stage["name"]] = check_cache(stage, cache_dir)
+
+        return {
+            "action": "show",
+            "stages": [
+                {
+                    "name": s["name"],
+                    "description": s["description"],
+                    "hash": s["hash"],
+                    "cached": cache_status.get(s["name"], False),
+                }
+                for s in stages
+            ],
+            "cache_stats": get_cache_stats(cache_dir),
+        }
+
+    elif action == "swap":
+        if not stage_name:
+            return {"error": "Stage name required for swap action"}
+        if not from_exp:
+            return {"error": "Source experiment ID required (--from)"}
+
+        experiments = load_experiments(log_path)
+        source = None
+        for exp in experiments:
+            if exp.get("experiment_id") == from_exp:
+                source = exp
+                break
+
+        if not source:
+            return {"error": f"Experiment {from_exp} not found"}
+
+        current_stages = define_stages(config)
+        new_stage = extract_stage_from_experiment(source, stage_name)
+        updated = swap_stage(current_stages, stage_name, new_stage)
+
+        return {
+            "action": "swap",
+            "stage": stage_name,
+            "source_experiment": from_exp,
+            "old_stage": next((s for s in current_stages if s["name"] == stage_name), None),
+            "new_stage": new_stage,
+            "updated_pipeline": [
+                {"name": s["name"], "description": s.get("description", ""), "hash": s.get("hash", "")}
+                for s in updated
+            ],
+        }
+
+    elif action == "cache":
+        stages = define_stages(config)
+        return {
+            "action": "cache",
+            "stages": [
+                {
+                    "name": s["name"],
+                    "hash": s["hash"],
+                    "cache_path": str(get_cache_path(s, cache_dir)),
+                    "already_cached": check_cache(s, cache_dir),
+                }
+                for s in stages
+            ],
+            "cache_dir": cache_dir,
+            "stats": get_cache_stats(cache_dir),
+        }
+
+    elif action == "run":
+        stages = define_stages(config)
+        skippable = [s["name"] for s in stages if check_cache(s, cache_dir)]
+        return {
+            "action": "run",
+            "total_stages": len(stages),
+            "cached_stages": skippable,
+            "stages_to_run": [s["name"] for s in stages if s["name"] not in skippable],
+            "message": f"Would skip {len(skippable)} cached stage(s) and run {len(stages) - len(skippable)}",
+        }
+
+    return {"error": f"Unknown action: {action}"}
+
+
+# --- Report Formatting ---
+
+
+def format_pipeline_report(report: dict) -> str:
+    """Format pipeline report as markdown."""
+    if "error" in report:
+        return f"ERROR: {report['error']}"
+
+    action = report.get("action", "?")
+    lines = [f"# Pipeline: {action.title()}", ""]
+
+    if action == "show":
+        lines.extend(["## Pipeline Stages", ""])
+        lines.append("| # | Stage | Description | Hash | Cached |")
+        lines.append("|---|-------|-------------|------|--------|")
+        for i, s in enumerate(report.get("stages", []), 1):
+            cached = "yes" if s.get("cached") else "no"
+            lines.append(f"| {i} | {s['name']} | {s['description']} | {s['hash']} | {cached} |")
+
+        stats = report.get("cache_stats", {})
+        if stats.get("total_files", 0) > 0:
+            lines.extend([
+                "",
+                f"**Cache:** {stats.get('total_files', 0)} files, "
+                f"{stats.get('total_size_mb', 0):.1f} MB",
+            ])
+
+    elif action == "swap":
+        old = report.get("old_stage", {})
+        new = report.get("new_stage", {})
+        lines.extend([
+            f"**Swapped stage:** {report.get('stage')}",
+            f"**Source experiment:** {report.get('source_experiment')}",
+            "",
+            f"Old: {old.get('description', 'N/A')} (hash: {old.get('hash', '?')})",
+            f"New: {new.get('config', {})} from {report.get('source_experiment')}",
+            "",
+            "## Updated Pipeline",
+            "",
+        ])
+        for s in report.get("updated_pipeline", []):
+            lines.append(f"- {s['name']}: {s['description']} ({s['hash']})")
+
+    elif action == "cache":
+        lines.append("| Stage | Hash | Path | Status |")
+        lines.append("|-------|------|------|--------|")
+        for s in report.get("stages", []):
+            status = "cached" if s.get("already_cached") else "to cache"
+            lines.append(f"| {s['name']} | {s['hash']} | {s['cache_path']} | {status} |")
+
+    elif action == "run":
+        cached = report.get("cached_stages", [])
+        to_run = report.get("stages_to_run", [])
+        lines.append(f"**{report.get('message')}**")
+        lines.append("")
+        if cached:
+            lines.append(f"Skip (cached): {', '.join(cached)}")
+        if to_run:
+            lines.append(f"Run: {', '.join(to_run)}")
+
+    return "\n".join(lines)
+
+
+def main() -> None:
+    """CLI entry point."""
+    parser = argparse.ArgumentParser(
+        description="Pipeline composition manager",
+    )
+    parser.add_argument(
+        "action", choices=["show", "swap", "cache", "run"],
+        help="Pipeline action",
+    )
+    parser.add_argument(
+        "stage", nargs="?",
+        help="Stage name (for swap)",
+    )
+    parser.add_argument(
+        "--from", dest="from_exp",
+        help="Source experiment ID (for swap)",
+    )
+    parser.add_argument(
+        "--config", default="config.yaml",
+        help="Path to config.yaml",
+    )
+    parser.add_argument(
+        "--log", default=DEFAULT_LOG_PATH,
+        help="Path to experiment log",
+    )
+    parser.add_argument(
+        "--cache-dir", default=DEFAULT_CACHE_DIR,
+        help=f"Cache directory (default: {DEFAULT_CACHE_DIR})",
+    )
+    parser.add_argument(
+        "--json", action="store_true",
+        help="Output raw JSON instead of formatted report",
+    )
+    args = parser.parse_args()
+
+    report = compose_pipeline(
+        action=args.action,
+        stage_name=args.stage,
+        from_exp=args.from_exp,
+        config_path=args.config,
+        log_path=args.log,
+        cache_dir=args.cache_dir,
+    )
+
+    if args.json:
+        print(json.dumps(report, indent=2, default=str))
+    else:
+        print(format_pipeline_report(report))
+
+
+if __name__ == "__main__":
+    main()

package/templates/scripts/scaffold.py

@@ -110,6 +110,9 @@ TEMPLATE_DIRS = {
         "experiment_diff.py",
         "training_monitor.py",
         "regression_gate.py",
+        "build_ensemble.py",
+        "pipeline_manager.py",
+        "warm_start.py",
     ],
     "tests": ["__init__.py", "conftest.py"],
 }
@@ -133,6 +136,9 @@ DIRECTORIES_TO_CREATE = [
     "experiments/diffs",
     "experiments/monitors",
     "experiments/regressions",
+    "experiments/ensembles",
+    "experiments/cache",
+    "experiments/warm_starts",
     "experiments/logs",
     "models/best",
     "models/archive",