Exercises

These exercises check the chapter's generalization ideas.

Exercise: Validation role

Enter 1 if validation loss is measured on training examples, or 2 if it is measured on held-out examples.

Your answer

Compute it first, then check your number.

Exercise: Underfit or overfit

Training loss is high and validation loss is high. Enter 1 for likely underfitting or 2 for likely overfitting.

Your answer

Compute it first, then check your number.

Exercise: Weight penalty

Let w = 5 and lambda = 0.2. What is lambda * w^2?

Your answer

Compute it first, then check your number.

Exercise: Dropout count

A layer has 50 activations. Dropout removes 20 in a training step. How many remain active?

Your answer

Compute it first, then check your number.

Exercise: Best epoch

Validation losses are [0.9, 0.6, 0.55, 0.7]. Which epoch has the lowest validation loss?

Your answer

Compute it first, then check your number.

Exercise: Augmented variants

Each of 40 examples is converted into 5 safe variants. How many variants are available?

Your answer

Compute it first, then check your number.

Exercise: Capacity ratio

A model has 2,400 parameters and 300 training examples. How many parameters per example is that?

Your answer

Compute it first, then check your number.