In the previous lesson,
you looked at using a real world data set, one with readings of sunspot activity
over a couple of hundred years. And you saw how to build a simple DNN to
predict activity using that data set. Now let's take a look at the notebook and
we'll see it in action for ourselves. As always, let's check if we
have TensorFlow 2 installed. And we do so we can just progress. This code just imports numpy and
pyplot and creates a helper function for plotting charts. This code downloads the data set and
puts it into the temp directory. It's a very small data set so
it should run very quickly. We'll now import the data from the CSV and
create numpy arrays from it. We can plot the data to
see it seasonality and its noise but there's no noticeable trend. We'll now split the data and
setup our constants. Most notably the window
size which I've set to 60. This code is a helper function for
turning the data into a Window data set. And now we'll train with a three layer
DNN having 2010, and one neurons. We'll use stochastic gradient
descent as our optimizer with a learning rate of 1e-7. When the training is done, we can get
forecasts across the validation set. And our plot looks pretty good. If we mentioned MAE, we're at about 15. You should get slightly different
values due to the randomness of the model initialization and
stochastic gradient descent. So don't worry if it's a little off. If it is way off, then go back and check your code or
restart the runtime and try again.