GPFLOW- Updating fitted model with a new datapoint without retraining from scratch (gaussian likelihood)

Question

I am using GPflow with gaussian likelihood for time series prediction. The data comes from a stream. Every time a new batch of data arrives (1 point per batch in my case), I want to update/extend my model with the new datapoint without retraining from scratch. Is this possible?

Right now I am retraining from scratch everytime. How can this be optimized? The running time of my model is of crutial importance so any help is welcome:

predictions = []  # To store the sequential predictions
model = GPR(data=(X_train, y_train), kernel=kernel)

for i in range(0, len(X_test), prediction_steps):
    
    # Train the model with the current training data
    optimizer = gpflow.optimizers.Scipy()
    start = time()
    optimizer.minimize(model.training_loss, model.trainable_variables)
    end = time()
    print('Time taken for iteration ', i, ' : ', end-start)
    # Make predictions for the next prediction_steps steps
    next_predictions, _ = model.predict_y(X_test[i : i + prediction_steps])

    # Store the predictions
    predictions.extend(next_predictions.numpy())

    # Update the training data with the true values for the next prediction_steps steps
    X_train = np.vstack([X_train, X_test[i : i + prediction_steps]])
    y_train = np.vstack([y_train, y_test[i : i + prediction_steps]])

    # Update the model's data
    model.data = (X_train, y_train)

(Each iteration yields the same running time)