Python Deep Learning - Second Edition

Example training

Now that we have our code for training and sampling, we can train the network on Leo Tolstoy's War and Peace and sample what the network has learned every couple of batch iterations. Let's prime the network with the phrase "She was born in the year" and see how it completes it during training.

The result we get after 500 batches is as follows: She was born in the year sive but us eret tuke Toffhin e feale shoud pille saky doctonas laft the comssing hinder to gam the droved at ay vime. The network has already picked up some distribution of characters and has come up with things that look like words.

After 5,000 batches, the network picks up a lot of different words and names: She was born in the year he had meaningly many ...