Syllabus Lesson 92 of 246 · Module 7: Working on new data
Module 7: Working on new data

How much data is enough?

Why it matters "Should we collect more data or build a bigger model?" is a question worth real money. A learning curve answers it.

Train the same model on growing slices of the data and plot validation accuracy against the number of examples. Early on, each doubling helps a lot. Later, each doubling helps less and less: the curve flattens. Where it has nearly flattened, more data will not help much and a better model might.

Call the Curve: from the accuracy at 50, 100 and 200 training digits, predict the accuracy at 400, then measure it.

Your turn

Write accuracy_with(n): train logistic regression on the first n training digits (pixels flattened) and return accuracy on the 400 test digits. Fill in my_guess_400 after looking at the first three values and before computing the fourth.

Spotted a problem in this lesson? Report it · Ask about this lesson

This exercise opens soon

Finish the lesson before it and this one unlocks. The whole course is free, no account needed, and you can read this lesson right now.