Rossman Sales Prediction - Deep Learning vs Random Forest


  1. Running lesson3-rossman notebook, I was able to reach an error of 0.103 on the test set using the DL approach, which is not too far from the 0.093 reported in the original paper, but not the same. Can this difference be attributed on randomization effects during training?

  2. More strangely, I tried to run the last few cells of the notebook which uses RF, and obtained an error of 0.109, which is much lower than 0.158 reported in the paper. Any hint on why this is that happening? Am I missing something?