Hands-On Neural Networks with Keras

Plotting training and test errors

In the following graph, we can see that the average error is about 2.5 (or $2,500 dollars). While this may be a small variance when predicting the prices on houses that cost $50,000, it starts to matter if the house itself costs $5,000:

Finally, let's predict some housing prices using data from the test set. We will use a scatter plot to plot the predictions and actual labels of our test set. In the following graph, we can see a line of best fit, along with the data points. Our model seems to capture the general trend in our data, despite having some outlandish predictions for some points:

Moreover, we can ...