Model optimization

[folguera]

Now that we have a working model, we should fine tune it

[cardinia]

Added early-stopping option:
7db47a7

Convergence was reached in only 3 epochs for the MPNN model, probably need to improve regularization via dropout layers and reducing the learning rate. When testing on a jupyter notebook the minimum was reached in 37 epochs with a learning rate of 0.005 and adding 0.2 dropout rate at each dense layer