import json from pathlib import Path import cloudpickle import numpy as np import pandas as pd class PacePredictor: def __init__(self, artifact_dir="."): self.artifact_dir = Path(artifact_dir) self.config = json.loads((self.artifact_dir / "model_config.json").read_text(encoding="utf-8")) with (self.artifact_dir / self.config["model_file"]).open("rb") as model_file: self.model = cloudpickle.load(model_file) self.feature_names = list(self.config["feature_order"]) self.threshold = float(self.config["decision_threshold"]) self.positive_class = self.config.get("positive_class", 1) classes = list(np.asarray(self.model.classes_)) if self.positive_class not in classes: raise ValueError("The configured positive class is missing from the fitted model.") self.positive_index = classes.index(self.positive_class) def predict(self, records): if isinstance(records, pd.DataFrame): frame = records.copy() elif isinstance(records, dict): frame = pd.DataFrame([records]) else: frame = pd.DataFrame(records) missing = [name for name in self.feature_names if name not in frame.columns] if missing: raise ValueError("Missing required model inputs: " + ", ".join(missing)) frame = frame.reindex(columns=self.feature_names) probabilities = self.model.predict_proba(frame)[:, self.positive_index] predictions = (probabilities >= self.threshold).astype(int) return pd.DataFrame({ "positive_class_probability": probabilities, "probability_percent": probabilities * 100.0, "predicted_label": predictions, "decision_threshold": self.threshold }, index=frame.index) def load_predictor(artifact_dir="."): return PacePredictor(artifact_dir=artifact_dir)