The checkerboard task is an extension to the logic XOR operation to continuous numbers. We'll treat this as a classification problem, by trying to predict the correct label (0 or 1) for each input.


You are given two versions of the analysis: One using pure TensorFlow and another using Keras as a high-level API on top of TensorFlow.

Tasks[edit | edit source]

  • Open the file 1_checkerboard/
    • Understand the classification ansatz (number of outputs, output activation, objective)
    • Train the network and evaluate the performance
    • Add another hidden layer
  • Open the file 1_checkerboard/
    • Compare with the TensorFlow syntax
    • Add another hidden layer
  • Play around network layout to see which layout gives a consistently good fit. Note, you can also do this on