Spaces:
Runtime error
Runtime error
File size: 12,423 Bytes
b91193c | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 | #!/usr/bin/env python3
"""Tune fusion weights + post-fusion modifiers; verify with live RankingPipeline."""
from __future__ import annotations
import argparse
import json
import logging
import subprocess
import sys
from datetime import datetime, timezone
from pathlib import Path
import optuna
PROJECT_ROOT = Path(__file__).resolve().parent.parent
sys.path.insert(0, str(PROJECT_ROOT))
from app.artifact_names import FUSION_PARAMS, MODIFIER_PARAMS, TUNING_REPORT_MODIFIERS # noqa: E402
from app.career_recall import load_career_scores # noqa: E402
from app.config import get_settings # noqa: E402
from app.fusion import ( # noqa: E402
build_fusion_cache,
evaluate_full_ranking,
settings_with_fusion_params,
)
from app.feature_store import load_silver_tiers # noqa: E402
from app.fusion import rank_from_fusion_cache # noqa: E402
from app.jd_requirements import load_or_build_jd_requirements # noqa: E402
from app.logging_setup import configure_logging # noqa: E402
from app.modifier_params import ModifierParams, save_modifier_params # noqa: E402
from app.pipeline import RankingPipeline # noqa: E402
from app.rank_data import load_candidates_lookup, load_feature_lookup # noqa: E402
logger = logging.getLogger(__name__)
POOL_SIZES = (500, 600, 700, 800)
CAP_SIZES = (28, 30, 32, 35)
CLONE_CAPS = (6, 8, 10, 12)
def _git_commit() -> str | None:
try:
out = subprocess.check_output(
["git", "rev-parse", "HEAD"],
cwd=PROJECT_ROOT,
stderr=subprocess.DEVNULL,
text=True,
)
return out.strip()
except (subprocess.CalledProcessError, FileNotFoundError):
return None
def verify_live_pipeline(
settings,
artifacts_dir: Path,
candidates_path: Path,
silver: dict[str, int],
candidate_lookup: dict,
) -> dict:
"""Run live RankingPipeline and evaluate same metrics as cache path."""
settings = settings_with_fusion_params(settings, artifacts_dir)
pipeline = RankingPipeline(settings, artifacts_dir)
results = pipeline.rank(candidates_path, top_k=100)
ranked_ids = [r["candidate_id"] for r in results]
return evaluate_full_ranking(
ranked_ids,
silver,
candidate_lookup,
reference_date=settings.reference_date,
)
def tune_modifiers(
settings,
artifacts_dir: Path,
candidates_path: Path,
*,
n_trials: int = 120,
max_rerank_pool: int = 800,
verify_live: bool = True,
) -> dict:
silver = load_silver_tiers(artifacts_dir / "labels_silver.parquet")
jd = load_or_build_jd_requirements(
settings.job_description_path,
artifacts_dir / "jd_requirements.json",
settings=settings,
)
feature_lookup = load_feature_lookup(artifacts_dir / "candidate_features.parquet")
candidate_lookup = load_candidates_lookup(candidates_path)
career_scores = load_career_scores(str(artifacts_dir))
cache = build_fusion_cache(settings, artifacts_dir, candidates_path, max_rerank_pool=max_rerank_pool)
def objective(trial: optuna.Trial) -> float:
ce_w = trial.suggest_float("rerank_ce_weight", 0.30, 0.55)
ltr_w = trial.suggest_float("rerank_ltr_weight", 0.45, 0.52)
rrf_w = 1.0 - ce_w - ltr_w
if rrf_w < 0.05 or rrf_w > 0.25:
raise optuna.TrialPruned()
mod = ModifierParams(
tier5_boost=trial.suggest_float("tier5_boost", 1.02, 1.12),
low_avail_mult=trial.suggest_float("low_avail_mult", 0.25, 0.45),
low_rr_steep_mult=trial.suggest_float("low_rr_steep_mult", 0.45, 0.65),
clone_tier5_mult=trial.suggest_float("clone_tier5_mult", 0.88, 0.98),
research_with_prod_mult=trial.suggest_float("research_with_prod_mult", 0.85, 0.98),
plain_language_career_boost=trial.suggest_float("plain_language_career_boost", 1.02, 1.12),
plain_language_career_threshold=trial.suggest_float("plain_language_career_threshold", 0.50, 0.65),
stretched_scientist_mult=trial.suggest_float("stretched_scientist_mult", 0.55, 0.75),
cv_speech_mult=trial.suggest_float("cv_speech_mult", 0.30, 0.50),
primary_hub_boost=trial.suggest_float("primary_hub_boost", 1.00, 1.10),
preferred_hub_boost=trial.suggest_float("preferred_hub_boost", 1.00, 1.06),
clone_top30_max=trial.suggest_categorical("clone_top30_max", list(CLONE_CAPS)),
top_availability_cap=trial.suggest_categorical("top_availability_cap", list(CAP_SIZES)),
)
pool_size = trial.suggest_categorical("rerank_pool_size", list(POOL_SIZES))
career_w = trial.suggest_float("career_rrf_weight", 0.50, 0.80)
trial_settings = settings.model_copy(update={"career_rrf_weight": career_w})
ranked = rank_from_fusion_cache(
cache,
jd=jd,
candidate_lookup=candidate_lookup,
feature_lookup=feature_lookup,
reference_date=trial_settings.reference_date,
rerank_pool_size=pool_size,
top_k=100,
ce_weight=ce_w,
ltr_weight=ltr_w,
rrf_weight=rrf_w,
modifier_params=mod,
career_scores=career_scores,
)
metrics = evaluate_full_ranking(
ranked,
silver,
candidate_lookup,
reference_date=trial_settings.reference_date,
)
if metrics["honeypots_in_top_10"] > 0 or metrics["low_avail_top_30"] > 0:
raise optuna.TrialPruned()
trial.set_user_attr("ndcg_at_10", metrics["ndcg_at_10"])
trial.set_user_attr("tier5_in_top_10", metrics["tier5_in_top_10"])
trial.set_user_attr("tier5_in_top_30", metrics["tier5_in_top_30"])
return (
metrics["ndcg_at_10"]
+ 0.02 * metrics["tier5_in_top_10"]
+ 0.008 * metrics["tier5_in_top_30"]
+ 0.003 * metrics["tier5_in_top_100"]
- 0.01 * metrics.get("template_clones_top_30", 0)
)
study = optuna.create_study(direction="maximize")
study.optimize(objective, n_trials=n_trials, show_progress_bar=False)
best = study.best_params
best["rerank_rrf_weight"] = round(1.0 - best["rerank_ce_weight"] - best["rerank_ltr_weight"], 4)
mod = ModifierParams(
tier5_boost=best["tier5_boost"],
low_avail_mult=best["low_avail_mult"],
low_rr_steep_mult=best["low_rr_steep_mult"],
clone_tier5_mult=best["clone_tier5_mult"],
research_with_prod_mult=best["research_with_prod_mult"],
plain_language_career_boost=best["plain_language_career_boost"],
plain_language_career_threshold=best["plain_language_career_threshold"],
stretched_scientist_mult=best["stretched_scientist_mult"],
cv_speech_mult=best["cv_speech_mult"],
primary_hub_boost=best["primary_hub_boost"],
preferred_hub_boost=best["preferred_hub_boost"],
clone_top30_max=best["clone_top30_max"],
top_availability_cap=best["top_availability_cap"],
)
cache_metrics = evaluate_full_ranking(
rank_from_fusion_cache(
cache,
jd=jd,
candidate_lookup=candidate_lookup,
feature_lookup=feature_lookup,
reference_date=settings.reference_date,
rerank_pool_size=best["rerank_pool_size"],
top_k=100,
ce_weight=best["rerank_ce_weight"],
ltr_weight=best["rerank_ltr_weight"],
rrf_weight=best["rerank_rrf_weight"],
modifier_params=mod,
career_scores=career_scores,
),
silver,
candidate_lookup,
reference_date=settings.reference_date,
)
fusion_payload = {
"rerank_ce_weight": best["rerank_ce_weight"],
"rerank_ltr_weight": best["rerank_ltr_weight"],
"rerank_rrf_weight": best["rerank_rrf_weight"],
"rerank_pool_size": best["rerank_pool_size"],
"career_rrf_weight": best["career_rrf_weight"],
"proxy_objective": study.best_value,
"cache_ndcg_at_10": cache_metrics["ndcg_at_10"],
"tier5_in_top_10": cache_metrics["tier5_in_top_10"],
"tier5_in_top_30": cache_metrics["tier5_in_top_30"],
"tier5_in_top_100": cache_metrics["tier5_in_top_100"],
"low_avail_top_30": cache_metrics["low_avail_top_30"],
"n_trials": n_trials,
"tuned_with_modifiers": True,
}
save_modifier_params(artifacts_dir / MODIFIER_PARAMS, mod)
(artifacts_dir / FUSION_PARAMS).write_text(json.dumps(fusion_payload, indent=2), encoding="utf-8")
live_metrics: dict | None = None
if verify_live:
settings = settings.model_copy(update={"career_rrf_weight": best["career_rrf_weight"]})
live_metrics = verify_live_pipeline(settings, artifacts_dir, candidates_path, silver, candidate_lookup)
fusion_payload["live_ndcg_at_10"] = live_metrics["ndcg_at_10"]
fusion_payload["live_tier5_in_top_10"] = live_metrics["tier5_in_top_10"]
fusion_payload["live_tier5_in_top_30"] = live_metrics["tier5_in_top_30"]
fusion_payload["live_tier5_in_top_100"] = live_metrics["tier5_in_top_100"]
(artifacts_dir / FUSION_PARAMS).write_text(json.dumps(fusion_payload, indent=2), encoding="utf-8")
save_modifier_params(
artifacts_dir / MODIFIER_PARAMS,
mod,
extra={
"live_ndcg_at_10": live_metrics["ndcg_at_10"],
"live_tier5_in_top_10": live_metrics["tier5_in_top_10"],
"live_tier5_in_top_30": live_metrics["tier5_in_top_30"],
"live_tier5_in_top_100": live_metrics["tier5_in_top_100"],
},
)
top10_ids = (
[r["candidate_id"] for r in RankingPipeline(settings_with_fusion_params(settings, artifacts_dir), artifacts_dir).rank(candidates_path, top_k=10)]
if verify_live
else rank_from_fusion_cache(
cache, jd=jd, candidate_lookup=candidate_lookup, feature_lookup=feature_lookup,
reference_date=settings.reference_date, rerank_pool_size=best["rerank_pool_size"], top_k=10,
ce_weight=best["rerank_ce_weight"], ltr_weight=best["rerank_ltr_weight"],
rrf_weight=best["rerank_rrf_weight"], modifier_params=mod, career_scores=career_scores,
)
)
report = {
"timestamp": datetime.now(timezone.utc).isoformat(),
"git_commit": _git_commit(),
"best_params": {**best, **{k: v for k, v in fusion_payload.items() if k.startswith("rerank")}},
"cache_metrics": cache_metrics,
"live_metrics": live_metrics,
"top10_fingerprint": top10_ids,
}
(artifacts_dir / TUNING_REPORT_MODIFIERS).write_text(json.dumps(report, indent=2), encoding="utf-8")
logger.info(
"Cache: NDCG@10=%.4f tier5@30=%.0f | Live: %s",
cache_metrics["ndcg_at_10"],
cache_metrics["tier5_in_top_30"],
f"NDCG@10={live_metrics['ndcg_at_10']:.4f}" if live_metrics else "skipped",
)
return fusion_payload
def main() -> int:
parser = argparse.ArgumentParser(description="Tune fusion + modifier params (full ranking path).")
parser.add_argument("--artifacts", type=Path, default=None)
parser.add_argument("--candidates", type=Path, default=None)
parser.add_argument("--trials", type=int, default=120)
parser.add_argument("--max-pool", type=int, default=800)
parser.add_argument(
"--verify-live",
action="store_true",
default=True,
help="Re-run live RankingPipeline after tuning (default: on).",
)
parser.add_argument("--no-verify-live", action="store_true", help="Skip live pipeline verification.")
args = parser.parse_args()
settings = get_settings()
configure_logging(settings.log_level)
artifacts_dir = (args.artifacts or settings.artifacts_dir).resolve()
candidates_path = (args.candidates or settings.candidates_path).resolve()
if not (artifacts_dir / "ltr_model.lgb").exists():
logger.error("Train LTR first: python scripts/train_ltr.py")
return 1
tune_modifiers(
settings,
artifacts_dir,
candidates_path,
n_trials=args.trials,
max_rerank_pool=args.max_pool,
verify_live=not args.no_verify_live,
)
return 0
if __name__ == "__main__":
raise SystemExit(main())
|