So let's now take
a look at some code for LSTMs and how they can work with the data that we've
been playing with all week. So here's the update to
our code to use LSTMs. Let's unpack it and take a look
at the interesting parts. First of all is the
tf.keras.backend.clear_session, and this clears
any internal variables. That makes it easy for
us to experiment without models impacting
later versions of themselves. After the Lambda layer that
expands the dimensions for us I've added a single LSTM
layer with 32 cells. I've also made a bidirectional to see the impact of
that on a prediction. The output neuron will give
us our prediction value. I'm also using a learning rate
of one times 10 _ minus six and that might be
worth experimenting with too. So here's the results
of running this LSTM on the synthetic data that we've been using throughout the course. The plateau under
the big spike is still there and are MAE is
in the low sixes. It's not bad, it's not
great, but it's not bad. The predictions look
like there might be a little bit on the low side too. So let's edit our code to add another LSTM to see the impact. Now you can see the second layer and note that we had to set return sequences equal to true on the first one in order
for this to work. We train on this and now we will see
the following results. Here's the chart.
Now it's tracking much better and closer
to the original data. Maybe not keeping up with the sharp increase but at
least it's tracking close. It also gives us a mean
average error that's a lot better and it's showing that we're heading in
the right direction. If we edit our source to add a third LSTM layer like this, by adding the layer and having the second layer
return sequences is true we can then
train and run it, and we'll get
the following output. There's really not that much of a difference and are MAE
has actually gone down. So that's it for looking at LSTMs and predicting
our sequences. In the next video
you'll take a look at using convolutions to see what the impact will be on using them for predicting
time-series content.