Common Beginner Confusions
Machine Learning has a few ideas that sound simple but are easy to misunderstand at first. Let's clear them up before moving forward.
The important thing is to understand the difference between data, training, a model, input, and prediction.
"Machine Learning means the computer thinks like a human."
A Machine Learning system does not need to understand the world the way a human does. It works with data and mathematical relationships to produce useful outputs.
A person can understand why a house might be expensive because of its location, condition, neighborhood, and many other factors.
Our example only knew about house size. It could not reason about information we never provided.
"The training data is the model."
Training data consists of examples. The learning process uses those examples to produce a model.
Examples
Find useful patterns
Learned representation
"Training and prediction are the same thing."
During training, the system learns from existing examples. During prediction, the trained model is used with new input.
Use training examples to learn a useful relationship.
Give the trained model new input and get an output.
"If I give new data to the model, it automatically learns."
A model can receive new data for prediction without changing what it has learned. Learning from new data requires an appropriate training or updating process.
Give the existing model a new house size and ask for a predicted price.
Use training data in a learning process to produce or update a model.
"More data always means a better model."
Adding useful and representative examples can help. Adding incorrect, irrelevant, or misleading examples can hurt.
Data that represents the problem accurately.
Incorrect, irrelevant, or misleading data.
"The model can use information even if I don't give it."
Our house-price example only used house size. Therefore, it could not directly use location, number of bedrooms, or house age.
"A prediction is guaranteed to be correct."
A model can make useful predictions while still making mistakes. Real-world data contains variation, noise, and information that may not be available to the model.
What the model estimates.
What actually happened.
A difference between the prediction and the actual value does not automatically mean the entire system is useless. We need appropriate evaluation methods to understand how well a model performs.
"Machine Learning is just writing Python code."
Python helps us implement Machine Learning algorithms, work with data, train models, and make predictions. The programming language itself is not Machine Learning.
The programming language we use.
The approach of learning useful patterns from data.
"Our simple house-price calculation is a complete Machine Learning model."
We intentionally calculated a very simple relationship ourselves so you could understand the core workflow without a Machine Learning library hiding the details.
Designed to make the basic idea easy to see.
We will learn algorithms such as Linear Regression properly.
Quick Reference
Training Data
Examples used during the learning process.
Training
The process of learning from training data.
Model
The learned representation used for future predictions or decisions.
Input
Information provided to the model.
Prediction
The output produced for new input.
Python
A programming language used to implement ML systems.
If you remember these five ideas, you're ready for the next lesson.
Machine Learning learns useful patterns from data.
Training and prediction are different stages.
A model represents what was learned.
Predictions depend on the model and the input provided to it.
The quality and relevance of training data matter.