Joseph Model

[jnnchi]

• Research any other ways we are able to train higher up the network–perhaps try out a higher degree of augmentation (50-70%), with two unfrozen layers and a low LR.
• Calculate F1 score to measure model accuracy
• Look into more ways we can improve our network. I don't necessarily think that means more “features” on top of the architecture we already have, but perhaps different ways to pass in data, etc.
• Feel free to try your model architecture using EfficientNet.