Deciding What to Do Next (Revisited)

  |   Source
Deciding What to Do Next (Revisited)
  • What to do or not to do to increase the learning algorightm

  • choosing small set of features when suffering High Bias, is NOT THE CORRECT WAY
  • All of the reaseon has discussed in previous videos
  • the left is less parameters and that is tend to have underfitting problem
  • Large Neural Networks is really prone to overfitting. Use regularization to fix this. Use that instead of using simpler neural neural networks.
  • number of hidden layers is your customize, but one hidden layer is the default.
  • Find cross validation, train set (split) and see which hidden layer number of units works best for neural networks.

  • That is for diagnose the curve from high variance/high bias
  • May all these tools and diagnostic can really help our learning algorithm to work really well