Grokking Algorithms

Introduction to machine learning
KNN is a really useful algorithm, and it’s your introduction to 
the magical world of machine learning! Machine learning is all 
about making your computer more intelligent. You already saw 
one example of machine learning: building a recommendations 
system. Let’s look at some other examples.
OCR
OCR stands for 
optical character recognition
. It means you can take a 
photo of a page of text, and your computer will automatically read the 
text for you. Google uses OCR to digitize books. How does OCR work? 
For example, consider this number.
How would you automatically figure out what number this is? You can 
use KNN for this: 
1. Go through a lot of images of numbers, and extract features of those 
numbers. 
2. When you get a new image, extract the features of that image, and 
see what its nearest neighbors are! 
It’s the same problem as oranges versus grapefruit. Generally speaking, 
OCR algorithms measure lines, points, and curves.
Then, when you get a new character, you can extract the same features 
from it.

---

200
Chapter 10
 
 
I
 
 
k-nearest neighbors
Feature extraction is a lot more complicated in OCR than the 
fruit example. But it’s important to understand that even complex 
technologies build on simple ideas, like KNN. You could use the same 
ideas for speech recognition or face recognition. When you upload a 
photo to Facebook, sometimes it’s smart enough to tag people in the 
photo automatically. That’s machine learning in action! 
The first step of OCR, where you go through images of numbers and 
extract features, is called 
training
. Most machine-learning algorithms 
have a training step: before your computer can do the task, it must
be trained. The next example involves spam filters, and it has a
training step.