API Reference

Complete reference for all public classes and functions in MaldiDeepKit, organized by module.

Base Classes

Every classifier inherits from BaseSpectralClassifier, which handles device placement, validation splits, early stopping, and persistence. Subclasses only need to override _build_model().

maldideepkit.BaseSpectralClassifier	Abstract base for all MaldiDeepKit classifiers.
maldideepkit.SpectralDataset	PyTorch Dataset wrapping a binned MALDI-TOF feature matrix.
maldideepkit.make_loaders	Build stratified train / validation DataLoader pairs.

Classifiers

maldideepkit.MaldiMLPClassifier	sklearn-compatible MLP classifier with optional attention gating.
maldideepkit.MaldiCNNClassifier	sklearn-compatible 1-D CNN classifier for MALDI-TOF spectra.
maldideepkit.MaldiResNetClassifier	sklearn-compatible 1-D ResNet classifier for MALDI-TOF spectra.
maldideepkit.MaldiTransformerClassifier	sklearn-compatible 1-D ViT classifier for MALDI-TOF spectra.

Building Blocks

Low-level nn.Module classes backing each classifier are re-exported through the maldideepkit.blocks namespace, so a user who wants to embed a single component in a custom network can import from one place. See Blocks Module for the full list, grouped into full backbones vs. composable primitives.

from maldideepkit.blocks import (
    SpectralTransformer1D,   # full backbones
    TransformerBlock,        # composable primitives
    PatchEmbed1D,
    BasicBlock1D,
    # ...
)

Training Utilities

maldideepkit.utils.seed_everything	Seed Python, NumPy, and PyTorch (CPU + CUDA) RNGs in one call.
maldideepkit.utils.resolve_device	Resolve a user-facing device specifier to a torch.device.
maldideepkit.utils.EarlyStopping	Track the best validation loss and signal when to stop training.
maldideepkit.utils.train_loop	Run a classic train + validate loop with early stopping.
maldideepkit.utils.FocalLoss	Multi-class focal loss with optional class weighting and label smoothing.
maldideepkit.utils.SAMOptimizer	Wrap a base optimizer in the SAM two-step update.
maldideepkit.utils.tune_threshold	Pick the binary decision threshold that maximises metric.
maldideepkit.utils.fit_temperature	Fit a scalar temperature by LBFGS minimisation of NLL.
maldideepkit.utils.find_lr	Sweep learning rate geometrically and return the LR / loss curve.
maldideepkit.utils.SpectralEnsemble	Ensemble N fitted or unfitted spectral classifiers.

Augmentation

Per-batch augmentations applied to training batches only; bypassed during validation and inference. See Augment Module for the full API.

maldideepkit.augment.SpectrumAugment	Composable per-batch spectrum augmentation.
maldideepkit.augment.apply_mixup	Mixup: convex-combine two random permutations of the batch.
maldideepkit.augment.apply_cutmix	CutMix on 1-D spectra: splice a contiguous m/z window.

Uncertainty Quantification

Drop-in estimators that wrap a fitted classifier and return calibrated predictions plus per-sample uncertainty. See Uncertainty Module.

maldideepkit.uncertainty.MCDropoutEstimator	Monte Carlo Dropout estimator (Gal and Ghahramani, 2016).
maldideepkit.uncertainty.LaplaceEstimator	Laplace-approximation uncertainty estimator.
maldideepkit.uncertainty.ConformalPredictor	Split conformal predictor with the LAC non-conformity score.
maldideepkit.uncertainty.UncertaintyResult	Container for the output of a BaseUncertaintyEstimator.