Lightning Modules¶

Lightning modules wrap model architectures with training logic. Each module handles loss computation, metric logging, optimizer configuration, and learning rate scheduling.

SegmentationModule¶

Binary segmentation of EPG signals. Uses Dice loss by default and logs Accuracy, Precision, Recall, F1, and IoU metrics.

from dnanet.modules.segmentation import SegmentationModule

Constructor args: - model — Any nn.Module producing (B, 1, D, L) output - loss_fn — Loss function (default: DiceLoss) - learning_rate — Initial learning rate - weight_decay — L2 regularization - scheduler_gamma — LR decay per epoch - threshold — Prediction threshold (default: 0.5)

ClassificationModule¶

Multi-class peak classification. Uses CrossEntropy or Focal loss.

from dnanet.modules.classification import ClassificationModule

ReconstructionModule¶

Autoencoder reconstruction. Uses MSE loss and logs reconstruction error.

from dnanet.modules.reconstruction import ReconstructionModule