13 steps for creating a Deep Learning model

Original article was published on Deep Learning on Medium


13 steps for creating a Deep Learning model

Well, This article might require minimum knowledge of how a deep learning model work. I have articulated the experiments I do once I create a baseline model. We always get stuck at the point ‘What’s next?!’. I haven’t found a blog explaining the step-by-step thought process from the scratch or experiment logs from people creating the model. So, here I am maintaining the logs from my experience for my reference which I thought would be useful for others looking for the same.

Ah! Where to go?! What to choose first?!

CHECKBOOK

  1. Check for Outliers
  2. Make a train-test data distribution test
  3. Find Feature Importance
  4. Make a one hidden layer NN model
  5. Choose an Initializer (Kernel & Bias)
  6. Check for Exploding Gradients
  7. Choose an activation function
  8. Choose an architecture of the model
  9. Find the stability in the predictions
  10. Find the optimal batch size
  11. Find the optimal epochs or the optimal ‘Patience’ if using Early Stopping
  12. Again, Find the stability in the predictions
  13. Check Performance metrics using Cross-Validation