scripts/run_pipeline_v4.py

#!/usr/bin/env python3
"""Pipeline: build_data + train_ranker (v4) — fixed HF downloads"""
import os, sys, gc, warnings
warnings.filterwarnings("ignore")
import numpy as np
import pandas as pd
from pathlib import Path

hf_token = os.environ.get("HF_TOKEN")
from huggingface_hub import HfApi, hf_hub_download
api = HfApi()

DS = "cedwyh/jinjing-shared-data"

print("=" * 60)
print("Pipeline: build_data + train_ranker (v4)")
print(" HF downloads via hf_hub_download (fixes urllib 404)")
print("=" * 60)

def hub_path(filename):
    """Get local path for a HF dataset file, downloading if needed"""
    return hf_hub_download(repo_id=DS, filename=filename, repo_type="dataset")

# ============================================================
# Step 1: Build ranking training data
# ============================================================
print("\n[1/2] Building ranking training data...")

build_data_local = hub_path("build_data.py")
# Patch factor_priors import
with open(build_data_local) as f:
    content = f.read()
content = content.replace(
    "from factor_priors import compute_all_priors, PRIOR_COLUMNS",
    "try:\n    from factor_priors import compute_all_priors, PRIOR_COLUMNS\nexcept ImportError:\n    compute_all_priors = None\n    PRIOR_COLUMNS = []"
)
patched = "/tmp/build_data_patched.py"
with open(patched, "w") as f:
    f.write(content)

sys.argv = ["build_data.py", "--output", "/tmp/ranking_train_v8.parquet",
            "--dataset", DS, "--no-use-priors"]
exec(open(patched).read().replace('if __name__ == "__main__"', 'if True'))

# Verify
if not Path("/tmp/ranking_train_v8.parquet").exists():
    print("\n❌ Build data failed - no output")
    sys.exit(1)

df = pd.read_parquet("/tmp/ranking_train_v8.parquet")
print(f"\n✅ Build data: {len(df):,} rows x {len(df.columns)} cols")
print(f"   Date range: {df['date'].min()} to {df['date'].max()}")
del df; gc.collect()

api.upload_file(
    path_or_fileobj="/tmp/ranking_train_v8.parquet",
    path_in_repo="ranking_train_v8.parquet",
    repo_id=DS, repo_type="dataset"
)
print("  ✅ Uploaded ranking_train_v8.parquet")

# ============================================================
# Step 2: Train LGBMRanker
# ============================================================
print("\n[2/2] Training LGBMRanker...")

train_ranker_local = hub_path("scripts/train_ranker.py")

output_dir = "/tmp/v10_ranker"
Path(output_dir).mkdir(parents=True, exist_ok=True)

sys.argv = ["train_ranker.py",
    "--data", "/tmp/ranking_train_v8.parquet",
    "--output", output_dir,
]

# Patch: the data has label_rank (1-10 decile), train_ranker.py expects label (binary)
# For lambdarank, label_rank works directly as ranking target
with open(train_ranker_local) as f:
    tr_content = f.read()
tr_content = tr_content.replace('TARGET_COL = "label"', 'TARGET_COL = "label_rank"')
tr_patched = "/tmp/train_ranker_patched.py"
with open(tr_patched, "w") as f:
    f.write(tr_content)

exec(open(tr_patched).read().replace('if __name__ == "__main__"', 'if True'))

# Upload models
model_files = sorted(Path(output_dir).glob("*.txt"))
pred_file = Path(output_dir) / "ranker_predictions.parquet"

print(f"\nModels: {len(model_files)}")
for f in model_files:
    api.upload_file(
        path_or_fileobj=str(f),
        path_in_repo=f"models/v10_{f.name}",
        repo_id=DS, repo_type="dataset"
    )
    print(f"  ✅ models/v10_{f.name}")

if pred_file.exists():
    api.upload_file(
        path_or_fileobj=str(pred_file),
        path_in_repo="models/ranker_v10_predictions.parquet",
        repo_id=DS, repo_type="dataset"
    )
    print("  ✅ Uploaded predictions")

print("=" * 60)
print("✅ Pipeline complete!")
print("   ranking_train_v8.parquet")
print("   models/v10_*.txt")
print("=" * 60)