In the last video,
we saw how to use RNNs for prediction of sequences, and we learned a
little about how to construct a neural
network with them. In this video,
you'll work through a notebook with all of the code, and when you're done
watching, you can try out the notebook
for yourself. As usual, let's first ensure
that we're running 2.0, and then we'll run
the code to set up the data and create the dataset. Then we'll run the neural
network code but adjusting the learning rate per epoch in order to find the best one
for the full training. We'll plot the results to try to find the optimal learning rate. You can see here, it's between
10_minus 6 and 10_minus 5. So I pick a value halfway say, 5 times 10_minus 5. Here's the code to train
the neural network. I've set the optimal
learning rates and I've picked 400 epochs
for which to train. I'll now train the neural
network for the 400 epochs. Once it's trained,
I can use it to forecast for the validation
range and plot the results. On all my plot, I can
see that my prediction isn't too bad other
than this plateau, which is definitely going to
impact my MA in a bad way. But despite that my MA is only about 6.41, so it's not too bad. If I plot the MA and loss
for training, I'll see this. Showing that while it looks quite flat after about epoch 50, it was actually still
reducing slowly, and when we zoom in, we can see that. We saw instability after
about epoch 400 earlier, so I just stopped
training at that point.