Lesson 1 of 6
What makes good training data
Explanation
Good training data is representative (it covers the real variety of cases the model will face), sufficient (enough examples to spot a pattern, not just noise), and correctly labeled (the "right answers" are actually right). A model trained only on daytime photos of cars will struggle at night — not because the algorithm is bad, but because the data never showed it that situation.
Visual explanation
Read this from left to right: what goes in, what happens, and what comes out.
The training examples need to cover many conditions before the model meets a real case.
In plain language
Good data looks like the real situations your model will face, has enough examples, and has correct answers.
Remember
Missing real-world situations create weak models.
What to do
A self-driving car model is trained only on sunny-day footage. What kind of data problem is this, and what would you add to fix it?