Skip to content

Latest commit

 

History

History
8 lines (8 loc) · 821 Bytes

imdb_reviews.md

File metadata and controls

8 lines (8 loc) · 821 Bytes
Raw
  • Get noteboook 3.5 from the book repo
  • Go through the notebook:
    • How the training time and accuracy change if you increase the size of middle layers from 16 to 32?
    • Does validation accuracy improve if you add another hidden layer?
    • In the loss and accuracy plots at the bottom the history object is used, explore that object. It contains information saved after each epoch of training.
    • Try usig and mse loss function instead of the binary_crossentropy
    • history.history['acc'] may fail for some of you, find what name replaces 'acc'. (Thie depends on how the network was compiled, there are 3 compilation examples in the notebook)
  • If done, train the same network on yelp data from yelp.ipynb from week 11 of the previous semester