In the previous video, you
looked at RNNs and how you could use RNNs to do
some simple prediction. You then improved on them, using LSTMs instead of RNNs to get a better fit
with our predicted data. In this video, you'll
take a look at the LSTM notebook after which you can try that notebook
out for yourself. As always, let's
ensure that we're using TensorFlow too,
before starting. And now that we see that we are, let's run this code
to create the data, and then the helper function that we'll turn it
into a dataset. Will then run our model, which has two LSTM layers. You can of course change this, but we'll run it to find
our optimum learning rate first. Once it's done, we can plot the loss versus
the learning rates, and we'll find
a good learning rate is about ten to the minus five. So, we'll update
our learning rates of that. I will train it for
about 500 epochs. When it's done, we'll
plot the forecast. As we can see, it's
definitely improved. Changing to LSTMs gets rid
of that plateau that we saw, and the curve is
closer to the data, but it's still not quite right, and it gives us an MAE
of a little over 8.5. So, that's a good start in using recurrent neural networks
to fix the problem. Have a play with a notebook and see if you can improve on it. Next week, we're going
to do one more thing, and that's adding
a convolutional layer before the LSTMs to see
the impact on training. After that, you'll
start moving away from synthetic data and you'll start using the real-world stuff.