Supervised Learning

Question 1

• What is being learned through model training and what is it learning from.

Answer 1

• Patterns, relationships, weights, parameters rules and functions which consists of features and labels. Learned from internet or any source of training data.

Question 2

• What is a prediction algorithm?

Answer 2

• Input to output

Question 3

• What are the different ways to judge the quality of a trained model?

Answer 3

Performance metrics (Confusion matrix, ROC and AUC curves, misfit)

Question 4

• Predictions on the training and the test datasets are called in sample and out-of-sample tests. Explain why both have a role in constructing a good model.

Answer 4

• In-sample – how good at calibrating
• Out-of-sample – how good at generalising

Question 5

• What trade-offs to think about when dividing our dataset into training and testing sets? Can this be done in more than one way?

Answer 5

• Training vs testing set size
• Yes, many ways such as random, time based or balanced

Question 6

• What factors to consider when selecting features for model training?

Answer 6

• How good the predictive skill on training set

Question 7

• What factors we consider when selecting a model to learn from the data?

Answer 7

• problem type, accuracy requirements, interpretability, complexity, scalability, feature space, and domain knowledge.

Question 8

• Supervision implies that the skilled operator is providing direction to the learning algorithm. What is this direction?

Answer 8

• Label data
• Defining objective
• Model performance
• Tuning algorithms
• Selecting features

Question 9

• Can the same structure be used to predict a binary label/target

Answer 9

Yes