"""Torch dataset and dataloader helpers.""" from __future__ import annotations from typing import List, Sequence import numpy as np import pandas as pd import torch from torch.utils.data import DataLoader, Dataset class AMPDataset(Dataset): def __init__( self, X: np.ndarray, samples: pd.DataFrame, target_cols: Sequence[str], mask_cols: Sequence[str], n_classes: int | None = None, ): self.X = np.array(X, dtype=np.float32, copy=True) self.samples = samples.reset_index(drop=True) self.target_cols = list(target_cols) self.mask_cols = list(mask_cols) self.n_classes = n_classes Y = np.array(self.samples[self.target_cols].to_numpy(dtype=np.float32), copy=True) M = np.array(self.samples[self.mask_cols].to_numpy(dtype=np.float32), copy=True) Y = np.nan_to_num(Y, nan=0.0) if n_classes is not None: Y = np.clip(Y, 0, n_classes - 1) self.Y = Y self.M = M def __len__(self) -> int: return len(self.samples) def __getitem__(self, idx: int): return { "x": torch.from_numpy(self.X[idx]), "y": torch.from_numpy(self.Y[idx]), "mask": torch.from_numpy(self.M[idx]), "idx": torch.tensor(idx, dtype=torch.long), } def make_loader( X: np.ndarray, samples: pd.DataFrame, target_cols: Sequence[str], mask_cols: Sequence[str], batch_size: int = 64, shuffle: bool = False, num_workers: int = 0, n_classes: int | None = None, ) -> DataLoader: ds = AMPDataset(X, samples, target_cols, mask_cols, n_classes=n_classes) return DataLoader( ds, batch_size=batch_size, shuffle=shuffle, num_workers=num_workers, pin_memory=torch.cuda.is_available(), drop_last=False, )