Skip to content

Callback — Base Training Callback

Base interface for observing and influencing MAGNET training loops.


Overview

Callback provides lifecycle hooks (on_train_begin/end, on_epoch_begin/end) that receive a mutable logs dictionary. Subclasses can inspect model state, trigger early stopping, or persist weights. EarlyStopping extends it with a concrete implementation.


Key Features

  • set_model() attaches the model before training
  • on_train_begin/end() hooks for full-training lifecycle
  • on_epoch_begin/end() hooks for per-epoch events
  • EarlyStopping subclass for automatic training termination

Methods

set_model(model)

Attach the current training model.

on_train_begin(logs=None)

Called once before training starts.

on_train_end(logs=None)

Called once after training finishes.

on_epoch_begin(epoch, logs=None)

Called at the beginning of each epoch.

on_epoch_end(epoch, logs=None)

Called at the end of each epoch; most common hook.


Example Usage

from machinegnostics.magnet import Callback

class PrintCallback(Callback):
    def on_epoch_end(self, epoch, logs=None):
        print(f"Epoch {epoch}: {logs}")

Author: Nirmal Parmar