2 Features, Labels, and Data Preparation Free Online FlashCards

What does each supervised-learning example pair?

A dataset is a collection of examples. In supervised learning, each example pairs features with a label.

What makes a feature suitable for prediction?

Features are the information given to a model for prediction. They should be useful and available when the model is used.

In dataset notation, what do $X$ and $y$ usually represent?

Features are often written as $X$, and labels as $y$.

What kinds of data can features contain?

Features can be numeric, categorical, text, dates, or other data types. The model may require transformations suited to its needs.

Name three ways to transform features during preparation.

Encoding categories, scaling numeric values, or converting dates into useful components can prepare features for a model.

What data-quality checks should precede training?

Check that examples match their labels, identify missing, invalid, or inconsistent values, and look for duplicates.

What is data leakage?

Data leakage occurs when features reveal the answer indirectly or include information unavailable at prediction time, making evaluation look overly good.

How should data-dependent preprocessing be fitted?

Fit data-dependent transformations using only the training set, then apply those fitted transformations to validation, test, and future data.

What are the training and validation sets used for?

The training set fits the model and learns patterns. The validation set supports development choices, such as comparing models or settings.

Why use stratified splitting for classification?

Stratified splitting can help preserve class proportions in each subset when creating classification splits.

How should data be split when multiple rows belong to one group?

Keep each person, household, or other group entirely within one subset so related rows do not cross split boundaries.

How should time-dependent data be split?

Split chronologically so training examples precede validation and test examples, reflecting how future data will be encountered.