Overview

General Ideas

• Learning from Data
• Identify Patterns
• Create Models
• Identifying features
• Predicting Targets

Models

• Learned functions that maps features to predictions
• Built (trained) by analyzing patterns in training data and adjusting internal parameters to minimize error
• Can be used on new, unseen data to make predictions

Core components of a model

• Parameters - e.g., weights in linear regression or or split points in decision trees
• Features (inputs) - variables the model uses to make predictions (e.g., location to predict house prices
• Outputs (targets/predictions) - result the model produces (e.g., predicted house price)

Types of machine learning models

• Supervised learning models
• Linear Regression
• Logistic Regression
• Tree-based models
• Support Vector Machines (SVMs)
• Neural Networks
• Unsupervised learning models
• Clustering models
• Dimensionality reduction models
• Reinforcement learning models
• Q-learning

Regression

Predicting a real number value

Example: Estimating Appartment Prices

Target: Price

Features: ?

Classification

Predicting a label

Example: Kind of Iris Plant

Target: Kind, eg. setosa, virginica or versicolor

Features: ?

Features and Targets

All entries must be numbers, i.e. all feature and target values must be transformed into numbers

Each feature vector is one instance/object

Visual Intro

A visual introduction to machine learning (link)

Model Tuning and the Bias-Variance Tradeoff (link)