Key Takeaways
You have now completed the first lesson. Before moving forward, let's reduce everything we learned into the ideas you should remember.
Machine Learning learns from examples and uses what it learned to make predictions or decisions.
That is the foundation behind everything we explored in this lesson.
ML learns patterns from data
Instead of manually writing every rule for every possible situation, we provide examples and use a learning process to find useful relationships in the data.
Examples provide the information
Training data contains examples that the learning process uses to find useful patterns or relationships.
A model represents what was learned
After learning from data, the resulting model can be used with new input to produce a prediction or decision.
Training and prediction are different
Training is the learning stage. Prediction is the stage where we use the trained model with new input.
The model depends on the data it learns from
If important training data changes, what the model learns can also change.
More importantly, the model cannot magically use information that was never provided to it.
Predictions are not guarantees
A model produces an estimate or decision based on what it learned. The prediction can be wrong, because real-world data is not perfectly predictable.
Python is a tool, not Machine Learning itself
We used Python to demonstrate the Machine Learning workflow. Later, we will use proper Machine Learning libraries and algorithms.
The Machine Learning Workflow
Examples
Learn patterns
What was learned
New example
Output
Everything in One Example
We started with houses for which we already knew the size and price.
Make Sure These Five Ideas Are Clear
ML learns useful patterns from data.
Training data is used during learning.
A model represents what was learned.
A trained model can make predictions for new inputs.
The data and information available to the model matter.
You now understand the basic Machine Learning workflow.
We started with a simple real-world problem, looked at data, found a relationship, created a simple model, made a prediction, changed the data, and examined what happened.
The next lesson will take this foundation and explain how real Machine Learning problems are structured.