TU Wien:Machine Learning VU (Musliu)/exam 2018 01 26

Exam 2018-01-26

There were 6 question blocks and in total 50 pts.

1. Yes- No Answers - 16 pts

• when k-nn is used with 1-n encoding, min-max scaling is needed to perform euclid. distance
• backpropagation is a technique used with perceptrons (single layer)
• pruning is needed for underfitting
• k-nn is recommended for large datasets
• paired t-tests are used when dealing with train/test splits
• for k-nn categorical features should be normalized

can't remember the other two

2. difference between z-score and min-max, when to use which on which feature types - 4 pts

3. difference naive bayesian network to normal network - 6 pts

4. difference gradient descent and normal equation. which would you use on a lin. reg. example - 6 pts

5. name some model based attributes in meta learning - 3 pts

6. example with 8 instances, perform 1R and naive bayes on 5 test examples (calculate accuracy, precision and recall), zero frequency was optional to handle - 7 pts 1R - 8 pts naive bayes