I want to train iteratively, first training on 0 and 1, but then pause training, modify the output layer to include a third output and then train on 0,1 AND 2 by adding 2s back in my dataset. I do not have to train from scratch. Train 10 epochs every time I add in a new digit. Just Be sure to train using the same model that was already trained with one less digit with the augmented output layer (the weights should not be reset only the output layer is increased by 1). After training for 10 epochs, add in 3 digits and expand the output layer of the network again. This has to be done all the way up until I have eventually done training on the entire dataset.
How to achieve this on the below code?