This short lesson summarizes key takeaways from this section.
You will be able to:
- Understand and explain what was covered in this section
- Understand and explain why this section will help you become a data scientist
The key takeaways from this section include:
- Probably Approximately Correct (PAC) learning theory provides a mathematically rigorous definition of what machine learning is
- The PAC is a learning model which is characterized by learning from examples
- Decision trees can be used for both categorization and regression tasks
- They are a powerful and interpretable technique for many machine learning problems (especially when combined with ensemble methods)
- Decision trees are a form of Directed Acyclic Graph (DAG) - you traverse them in a specified direction, and there are no "loops" in the graphs to go backwards
- Algorithms for generating decision trees are designed to maximize the information gain from each split
- A popular algorithm for generating decision trees is ID3 - the Iterative Dichotomiser 3 algorithm
- There are a range of pruning hyperparameters for decision trees to reduce overfitting - including maximum depth, minimum samples leaf with split, minimum leaf sample size, maximum lead nodes and maximum features
- CART (Classification and Regression Trees) trees can be used for regression tasks