predictor.py

import logging
import pandas as pd
from typing import List

from models.artifacts_loader import get_artifacts
from schemas.boq import BoQRequest, BoQResponse, BoQPrediction
from core.exceptions import ModelPredictionError
from utils.inference import predict_with_cascade

logger = logging.getLogger(__name__)


def _safe_float(value) -> float | None:
    """
    Safely coerce a value to float, returning None for any uncoercible input.

    Handles Excel overflow artifacts like '########' (column too narrow to
    display the number), empty strings, and any other non-numeric garbage
    that may arrive from poorly-formatted spreadsheets.
    """
    if value is None:
        return None
    if isinstance(value, float) and pd.isna(value):
        return None
    try:
        return float(value)
    except (ValueError, TypeError):
        return None


def predict_boq_batch_service(request: BoQRequest) -> BoQResponse:
    """
    Core service logic to predict NRM categories for a batch of BoQ items.
    """
    model = get_artifacts().model
    if not model:
        raise ModelPredictionError("ML model is currently unavailable.")

    fallback_engine = get_artifacts().fallback_engine
    if not fallback_engine:
        raise ModelPredictionError("Fallback engine is currently unavailable.")

    try:
        # 1. Convert JSON payload to DataFrame
        df_raw = pd.DataFrame(request.items)

        # 2. Delegate EVERYTHING to the Hermetic Production Object and Fallback Engine
        df_results = predict_with_cascade(
            df_raw=df_raw, ml_model=model, fallback_engine=fallback_engine
        )

        # 3. Format the response
        predictions_list: List[BoQPrediction] = []
        for _, row in df_results.iterrows():
            predictions_list.append(
                BoQPrediction(
                    predicted_category=str(row.get("Predicted_Category", "Unknown")),
                    confidence=float(row.get("Prediction_Confidence", 0.0)),
                    prediction_source=str(row.get("Prediction_Source", "ML_Model")),
                    
                    predicted_ge=str(row["Predicted_ge"]) if "Predicted_ge" in df_results.columns else str(row.get("ge", "")),
                    predicted_group_element=str(row["Predicted_group_element"]) if "Predicted_group_element" in df_results.columns else str(row.get("group_element", "")),
                    predicted_e=str(row["Predicted_e"]) if "Predicted_e" in df_results.columns else str(row.get("e", "")),
                    predicted_element=str(row["Predicted_element"]) if "Predicted_element" in df_results.columns else str(row.get("element", "")),
                    
                    ge=str(row["ge"]) if "Predicted_ge" in df_results.columns and pd.notna(row["ge"]) else None,
                    group_element=str(row["group_element"]) if "Predicted_group_element" in df_results.columns and pd.notna(row["group_element"]) else None,
                    e=str(row["e"]) if "Predicted_e" in df_results.columns and pd.notna(row["e"]) else None,
                    element=str(row["element"]) if "Predicted_element" in df_results.columns and pd.notna(row["element"]) else None,
                    
                    description=str(row.get("description", "Unknown")),
                    item=str(row.get("item")) if pd.notna(row.get("item")) else None,
                    notes=str(row.get("notes")) if pd.notna(row.get("notes")) else None,
                    qty=_safe_float(row.get("qty")),
                    rate_base_date=_safe_float(row.get("rate_base_date")),
                    rate_q1_2025=_safe_float(row.get("rate_q1_2025")),
                    uom=str(row.get("uom")) if pd.notna(row.get("uom")) else None,
                    project=str(row.get("project")) if pd.notna(row.get("project")) else None,
                    typology=str(row.get("typology")) if pd.notna(row.get("typology")) else None,
                    location=str(row.get("location")) if pd.notna(row.get("location")) else None,
                    base_date=str(row.get("base_date")) if pd.notna(row.get("base_date")) else None,
                    package=str(row.get("package")) if pd.notna(row.get("package")) else None,
                    rate_scope=str(row.get("rate_scope")) if pd.notna(row.get("rate_scope")) else None,
                    client_location=str(row.get("client_location")) if pd.notna(row.get("client_location")) else None,
                    basket_of_goods=str(row.get("basket_of_goods")) if pd.notna(row.get("basket_of_goods")) else None
                )
            )

        return BoQResponse(
            model_version=model.model_version,
            model_name=model.model_name,
            predictions=predictions_list,
        )

    except Exception as e:
        logger.error(f"Inference error in service: {e}", exc_info=True)
        raise ModelPredictionError(f"Internal inference error: {str(e)}") from e