My parity plot is horizontal, how do I fix that?

If your parity plots look something this like (i.e., constant DP predictions):

You should try decreasing the initial learning rate (start_lr in the in.json training parameter file).

For example, changing start_lr=0.001 (above image) to start_lr=0.0003 resulted in:

Success! But you’ll need to play around with it for your system(s).