Let's take a look at
training cats versus dogs with a smaller dataset
over a longer cycle. I'm going to start
the training and we'll see it picked up the 2,000
training images, but note that we're
training for 100 epochs. After the first epoch, We'll watch the accuracy
and validation accuracy figures. Let's watch them for
a few more epochs. Keep an eye on those figures. After eight epochs, the accuracy
is approaching 0.8, but the validation accuracy
has slowed its growth. So now let's skip
ahead to the end. I'm going to plot
the accuracy and loss overall 100 epochs. We can see from this figure that the training reach close to a 100 percent accuracy in
a little over 20 epochs. Meanwhile, the validation topped out at around 70 percent, and that's overfitting
clearly been demonstrated. In other words,
the neural network was terrific at finding
a correlation between the images and labels
of cats versus dogs for the 2,000 images
that it was trained on, but once it tried to
predict the images that it previously hadn't seen, it was about 70 percent accurate. It's a little bit
like the example of the shoes we
spoke about earlier. So in the next video, we'll take a look
at the impact of adding augmentation to this.