6. Model Devleopment

[JoeyTaubert]

Modify RNN

• Decide what data type we are going to be working with:
  • computed metrics (int/float) or raw packet data (string of raw bytes)
• Clean up error handling to reduce panics (maybe save this for Sprint 4)
• Feed info from my capture files directly into the network inputs

Add LSTM concepts

• Input Gate
• Forget Gate
• Output Gate
• Careful Initialization of Weights
  • As opposed to random initialization
• New Training Function/Modify Training Function
• New Activation Function (ReLU or Softsign)
• Set Dropout Rates for Connections Between LSTM Cells
• New State Vectors for Hidden State and Cell State