notes.md

• Try various hidden layer sizes
• Try different dataset
• Try randomizing the w2_random for each batch or epoch.
  • Tried it for each batch and the network was not learning
  • Also tried it for each epoch. This is slightly less bad than doing it for each batch