How to use the medaka.util.generators.serve_sample_batch function in medaka

'(reference gives correct base).'.format(train_benchmark))
    print('benchmark accuracy on validation set {}, '
          '(reference gives correct base).'.format(validate_benchmark))
    print('benchmark accuracy on test set {}, '
          '(reference gives correct base).'.format(test_benchmark))

    np.random.seed(0)

    # setup model
    feature_length = X_train.shape[1]
    batch_shape = (batch_size, window_size, feature_length)
    model = build_lstm(batch_shape)
    model.summary()

    # setup data generators
    train_generator = serve_sample_batch(X_train, y_train, *params)
    validate_generator = serve_sample_batch(X_validate, y_validate, *params)
    test_generator = serve_sample_batch(X_test, y_test, *params)

    # setup callbacks
    training_outfile = '_'.join([out_prefix, 'training_history.txt'])
    csv_logger = CSVLogger(training_outfile, separator='\t')
    checkpoint_out = '_'.join([out_prefix, 'checkpoint_model.epoch{epoch:02d}.h5'])
    model_checkpoint = ModelCheckpoint(checkpoint_out)
    callbacks = [csv_logger, model_checkpoint]

    # train model
    model.fit_generator(train_generator, steps_per_epoch=train_steps,
                        epochs=epochs, validation_data=validate_generator,
                        validation_steps=validate_steps, callbacks=callbacks,
                        workers=threads, class_weight=class_weights)

print('benchmark accuracy on validation set {}, '
          '(reference gives correct base).'.format(validate_benchmark))
    print('benchmark accuracy on test set {}, '
          '(reference gives correct base).'.format(test_benchmark))

    np.random.seed(0)

    # setup model
    feature_length = X_train.shape[1]
    batch_shape = (batch_size, window_size, feature_length)
    model = build_lstm(batch_shape)
    model.summary()

    # setup data generators
    train_generator = serve_sample_batch(X_train, y_train, *params)
    validate_generator = serve_sample_batch(X_validate, y_validate, *params)
    test_generator = serve_sample_batch(X_test, y_test, *params)

    # setup callbacks
    training_outfile = '_'.join([out_prefix, 'training_history.txt'])
    csv_logger = CSVLogger(training_outfile, separator='\t')
    checkpoint_out = '_'.join([out_prefix, 'checkpoint_model.epoch{epoch:02d}.h5'])
    model_checkpoint = ModelCheckpoint(checkpoint_out)
    callbacks = [csv_logger, model_checkpoint]

    # train model
    model.fit_generator(train_generator, steps_per_epoch=train_steps,
                        epochs=epochs, validation_data=validate_generator,
                        validation_steps=validate_steps, callbacks=callbacks,
                        workers=threads, class_weight=class_weights)

    # report final results on test data

'(reference gives correct base).'.format(validate_benchmark))
    print('benchmark accuracy on test set {}, '
          '(reference gives correct base).'.format(test_benchmark))

    np.random.seed(0)

    # setup model
    feature_length = X_train.shape[1]
    batch_shape = (batch_size, window_size, feature_length)
    model = build_lstm(batch_shape)
    model.summary()

    # setup data generators
    train_generator = serve_sample_batch(X_train, y_train, *params)
    validate_generator = serve_sample_batch(X_validate, y_validate, *params)
    test_generator = serve_sample_batch(X_test, y_test, *params)

    # setup callbacks
    training_outfile = '_'.join([out_prefix, 'training_history.txt'])
    csv_logger = CSVLogger(training_outfile, separator='\t')
    checkpoint_out = '_'.join([out_prefix, 'checkpoint_model.epoch{epoch:02d}.h5'])
    model_checkpoint = ModelCheckpoint(checkpoint_out)
    callbacks = [csv_logger, model_checkpoint]

    # train model
    model.fit_generator(train_generator, steps_per_epoch=train_steps,
                        epochs=epochs, validation_data=validate_generator,
                        validation_steps=validate_steps, callbacks=callbacks,
                        workers=threads, class_weight=class_weights)

    # report final results on test data
    loss_and_metrics = model.evaluate_generator(test_generator, test_steps,