Knowledge check

Check your knowledge

What does it mean to have an imbalanced dataset?

The number of samples is smaller or greater than is required

The feature columns contain many missing values

There are many more training examples that correspond to some outputs (categories) than others

What information can we extract from a single confusion matrix?

Log loss, and/or mean squared error

Whether the dataset has overfitted the training set

What kind of mistakes the model is making

Why are measures like True Positives or Accuracy not used to train our models directly?

There are mathematical barriers that prevent these being used for some training regimens

Subtle model improvements often don't affect these metrics

Both of the above

You must answer all questions before checking your work.