Vocabulary
- Instance based model: Requires all training data to be stored to work
- Hyperparameter: Argument to the model that determines how the model behaves
- Model: a mathematical object that
- takes your training data,
- learns a function mapping data to labels,
- and then is able to take unlabeled data and assign it labels
Classification Problems
A type of supervised machine learning
Goal: Assign categories or labels to data points based on patterns in the data
- There’s a target variable you want to predict
- You have historical data where the target label is known
- You have new data where the target variable is unknown
Data Splits
Data is split into three categories:
- Training data
- Testing data
- Validation data
Types of Learning
- Supervised learning: Learns from labeled data where each example in training set has known target or output
- Classification: Target is discrete
- Regression: Target is continuous
- Unsupervised learning: Learns from unlabeled data and tries to identify patterns or structures within data
- e.g. clustering similar points or reducing dimensionality
- Reinforcement learning: Learns to make sequence of decisions to maximize reward
- Uses rewards and punishments
- Used in robotics, games, autonomous systems
Subtopics
- Feature Engineering
- Evaluation
- Classification
- Regression
- Neural Networks
- Debugging models
- Graphs (as in graph theory)