Machine Learning Crash Course: Core Concepts
The essential vocabulary of machine learning: features, labels, models, training, and loss.
Inspired by learning material from Google
Machine learning lets computers learn from data. Here are the core ideas you'll meet everywhere.
Features and labels
- A feature is an input variable (e.g. square footage of a house).
- A label is the answer we want to predict (e.g. the house price).
The model
A model is a mathematical function that maps features to a prediction. Training is the process of finding the best parameters for that function.
Loss
Loss measures how wrong the model's predictions are. Training works by gradually reducing loss using an algorithm called gradient descent.
Training, validation, and test sets
- Training set: data the model learns from.
- Validation set: used to tune settings.
- Test set: held back to measure real performance.
Overfitting
If a model memorizes the training data instead of learning general patterns, it overfits and performs poorly on new data. We fight this with more data, simpler models, and regularization.
Key takeaways
- Models learn a mapping from features to labels.
- We minimize loss to improve predictions.
- Always evaluate on data the model hasn't seen.