You have a dataset with 20 classes.

How many classifiers will you have to train when you use the one versus one strategy to train a linear SVM?

You have a dataset with 20 classes.

How many classifiers will you have to train when you use the one versus all strategy to train a linear SVM?

Among the following properties, which are the required ones for a test set?

When training a linear SVM model, the C meta-parameter weights the cost of classification errors (versus the margin).

What happens if you increase the value of C meta-parameter?

To limit the influence of noise (erroneous samples) when training a k-nearest neighbor classifier, what should you do?

Among the following properties, which ones are true regarding decision trees?

You have a dataset with 20 classes.

How many classifiers will you have to train when you use the one versus all strategy to train a decision tree?

Select the problem each approach can solve.

Linear regression is sensitive to noise (outliers in training data) because of what?

What is the simplest clustering technique (hence the one you should use) to cluster a distribution like this?