This was a fun one! The videos/cnn-text example got a bit mangled: one of the examples was using a different version of the dataset (downloaded from Keras) than the other (using a hand-rolled downloader), and the hand-rolled downloader was missing!
I rewrote it (download-imdb.py) and the data loading code (imdb.py) to remove some bugs and do my best to make them OS-independent.
Also, the differences in setup meant that the code had been changed since the video. I reverted it and confirmed that we could get good performance.
This was a fun one! The
videos/cnn-textexample got a bit mangled: one of the examples was using a different version of the dataset (downloaded from Keras) than the other (using a hand-rolled downloader), and the hand-rolled downloader was missing!I rewrote it (
download-imdb.py) and the data loading code (imdb.py) to remove some bugs and do my best to make them OS-independent.Also, the differences in setup meant that the code had been changed since the video. I reverted it and confirmed that we could get good performance.