"""Prediction, evaluation, and submission helpers.""" from __future__ import annotations from pathlib import Path from typing import Any, Dict, Optional import numpy as np import pandas as pd import torch from ..utils import save_json from .decision import point_from_output from .metrics import aggregate_metrics, prediction_frame @torch.no_grad() def predict_loader( model: torch.nn.Module, loader, device: torch.device, output_type: str = "regression", n_classes: int = 87, target_scaler=None, ) -> tuple[np.ndarray, np.ndarray, np.ndarray]: model.eval() preds = [] targets = [] masks = [] for batch in loader: x = batch["x"].to(device, non_blocking=True) pred = model(x) point = point_from_output(pred, output_type, n_classes) if target_scaler is not None and output_type == "regression": point = torch.from_numpy(target_scaler.inverse(point.cpu().numpy())).to(device).float() preds.append(point.cpu()) targets.append(batch["y"]) masks.append(batch["mask"]) return ( torch.cat(preds).numpy(), torch.cat(targets).numpy(), torch.cat(masks).numpy(), ) def evaluate_predictions( pred_df: pd.DataFrame, metrics=("smape_plus_1", "mae"), bootstrap: int = 0, seed: int = 42, ) -> Dict[str, Any]: return aggregate_metrics( pred_df, metrics=metrics, per_target=True, per_horizon=True, per_cohort=True, per_fold=True, bootstrap=bootstrap, seed=seed, ) def save_predictions(pred_df: pd.DataFrame, path: str | Path) -> None: p = Path(path) p.parent.mkdir(parents=True, exist_ok=True) pred_df.to_csv(p, index=False) def make_submission( pred_df: pd.DataFrame, rating_col: str = "y_pred", group_key_col: str = "group_key", clip_min: float = 0.0, clip_max: float = 100.0, ) -> pd.DataFrame: df = pred_df.copy() df["prediction_id"] = ( df["visit_id"].astype(str) + "_" + df["target"].astype(str) + "_plus_" + df["horizon"].astype(int).astype(str) + "_months" ) df["rating"] = df[rating_col].clip(clip_min, clip_max) if group_key_col not in df.columns: df[group_key_col] = df["visit_month"] return df[["prediction_id", "rating", group_key_col]].rename(columns={group_key_col: "group_key"}) def save_metrics(metrics: Dict[str, Any], path: str | Path) -> None: save_json(metrics, path)