| import json |
| import joblib |
| import pandas as pd |
| from huggingface_hub import hf_hub_download |
|
|
| BERKAS = ["model_gb_tuned.joblib", "scaler.joblib", "model_columns.json", "preprocessing_config.json"] |
|
|
|
|
| def muat_model(repo_id): |
| p = {nama: hf_hub_download(repo_id=repo_id, filename=nama) for nama in BERKAS} |
| return { |
| "model": joblib.load(p["model_gb_tuned.joblib"]), |
| "scaler": joblib.load(p["scaler.joblib"]), |
| "columns": json.load(open(p["model_columns.json"])), |
| "config": json.load(open(p["preprocessing_config.json"])), |
| } |
|
|
|
|
| def siapkan_fitur(data_mentah, bundel): |
| cfg = bundel["config"] |
| df = data_mentah.copy() if isinstance(data_mentah, pd.DataFrame) else pd.DataFrame(data_mentah) |
| df = df[cfg["urutan_fitur_input"]] |
| for kol, valid in cfg["kategori_valid"].items(): |
| asing = set(df[kol].astype(str)) - set(valid) |
| if asing: |
| raise ValueError(f"Nilai {sorted(asing)} pada kolom '{kol}' tidak dikenal model. Nilai valid: {valid}") |
| X = pd.get_dummies(df, columns=cfg["fitur_kategorikal"]) |
| X = X.reindex(columns=bundel["columns"], fill_value=0).astype(float) |
| if cfg.get("model_membutuhkan_scaling"): |
| X[cfg["fitur_numerik"]] = bundel["scaler"].transform(X[cfg["fitur_numerik"]]) |
| return X |
|
|
|
|
| def prediksi_gaji(data_mentah, bundel): |
| |
| return bundel["model"].predict(siapkan_fitur(data_mentah, bundel)) |
|
|
