Let's take a deeper look into our results. In particular, we would like to know what kind of images our CNN does well in, and what kind of images it gets wrong.
Recall that the output of the sigmoid activation function in the last layer of our CNN is a list of values between 0 and 1 (one value/prediction per image). If the output value is < 0.5, then the prediction is class 0 (that is, cat) and if the output value is >= 0.5, then the prediction is class 1 (that is, dog). Therefore, an output value close to 0.5 means that the model isn't so sure, while an output value very close to 0.0 or 1.0 means that the model is very sure about its predictions.
Let's run through the images in the testing set one by one, using our model to make predictions on the class of the image, and classify the images according to three categories:
- Strongly right...