Create a better auto-naming policy for the model/weight output

[RobertMcCoy]

Should auto-concatenate the epoch count and have some other unique identifiers that represent what the model truly means - maybe find a way to get a count of how many convolutional layers and auto-interpolate that to the string?