So now, let's take a look at
how to get a prediction for a word and how to generate new text based on
those predictions. So let's start with
a single sentence. For example, 'Lawrence
went to Dublin.' I'm calling this
sentence the seed. If I want to predict
the next 10 words in the sentence to follow this, then this code will
tokenizer that for me using the text to sequences
method on the tokenizer. As we don't have
an outer vocabulary word, it will ignore 'Lawrence,'
which isn't in the corpus and will get
the following sequence. This code will then
pad the sequence so it matches the ones
in the training set. So we end up with something like this which we can pass to the model to
get a prediction back. This will give us
the token of the word most likely to be
the next one in the sequence. So now, we can do a reverse lookup on the word
index items to turn the token back into
a word and to add that to our seed texts, and that's it. Here's the complete
code to do that 10 times and you can
tweak it for more. But do you know that
the more words you predict, the more likely you are
going to get gibberish? Because each word is predicted, so it's not 100 per cent certain, and then the next one
is less certain, and the next one, etc. So for example, if you try the same seed and
predict 100 words, you'll end up with
something like this. Using a larger corpus we'll help, and then the next video, you'll see the impact of that, as well as some tweaks
that a neural network that will help
you create poetry.