Goal is to create a Neural Network that will predict soundspeed at multiple depths using spectrograms created from sound of ship passages recorded on sea floor. Currently using a mixed data structure with a convolutional net as one branch and enviormental and ship data as a MLP as the other branch.
Models structures are defined in the JSON directory and can be remade immediately. A method to access trained models is on the way. Training code can be found in the ml_attempts directory in general the longer the name the more likely its a more recent attempt. Most of the data preprocessing code is in the base directory I'm going to clean it all up into a better structure soon though.