Hands-On Machine Learning for Algorithmic Trading

Loading the IMDB movie review data

To keep the data manageable, we will illustrate this use case with the IMDB reviews dataset, which contains 50,000 positive and negative movie reviews evenly split into a train and a test set, and with balanced labels in each dataset. The vocabulary consists of 88,586 tokens.

The dataset is bundled into Keras and can be loaded so that each review is represented as an integer-encoded sequence. We can limit the vocabulary to num_words while filtering out frequent and likely less informative words using skip_top, as well as sentences longer than maxlen. We can also choose oov_char, which represents tokens we chose to exclude from the vocabulary on frequency grounds, as follows:

from keras.datasets import imdb ...