Testing the model

Welcome back.

As you can see we are separating the test in a different lecture.

That’s because we want to encourage you to do it on your own as homework if you want to try that Please

pause the video download the file associated with the lesson and go for it or you can skip it for now

and proceed with a lecture to test the model we use the method evaluate let’s declare two variables

Test loss and test accuracy equal to model evaluate of the test inputs and test targets.

And that’s about it.

Recall that evaluate returns the loss and any other metrics we’ve requested in our model outline in

our case.

That was the accuracy to make the result pretty.

We can print them with some nice formatting.

Good.

That’s the final accuracy of the model.

Naturally it is close to the validation accuracy as we did not fiddle too much with hyper parameters.

Note that sometimes you can get a test accuracy higher than the validation one that’s nothing but pure

luck.

Theoretically the test accuracy should be lower or equal to the validation one.

Okay from this point on I am no longer allowed to change the model.

I’ll follow this simple machine learning rule and let you do the hard work too.

So there are several exercises waiting for you place jump ahead and give them a go.

Good luck and thanks for watching.