Part of course:

Perceptron algorithm

- The perceptron
- The learning algorithm
- Why does the algorithm work?
- Algorithm at work:
- Notes:

NaN.

Perceptron algorithm[ Edit ]

Lets say you have a dataset for binary classification. Each data point x has n-dimensions, and the target value is y = 1 or -1 for the two classes.

The perceptron algorithm can be visualized as follows.

For a given data point x, the prediction is simply based on whether the linear function f(x) = b + w_{1}*x_{1} + w_{2}*x_{2} + ... + w_{n}*x_{n} is greater than or lesser than 0 (in the above diagram, w_{0} is b).

define predict(x):let f(x) := dot_product(x, w) + bprediction = 1 if f(x) > 0 and -1 if f(x) < 0

The learning algorithm is iterative. We start by initializing w with random values, and iteratively update its value.

define perceptron():A: initialize weights w randomly (n dimensions)B: for each data point (x, y) in the datasetmake prediction p given input xif p is not equal to y (i.e. prediction is wrong)w = w + y*x (update the weights)C: repeat step B until no points are misclassified

The algorithm works because if a point is misclassified, then the new weights are *better* for the point than the old weights. [1]

For example, suppose for the data point (x, y) that f(x) < 0 and y = 1. Then, after the update, we would want f(x) to be higher than before the update (so that after many updates, it becomes > 0).

Old weights = w, and new weights = w + x. It is easy to verify that

dot-product(w + x, x) + b > dot-product(w, x) + b

Here is an animation of the algorithm during training. Notice how the line moves only when a point is misclassified.

- Sometimes people initialize the weights to 0 instead of with random values (algorithm step A).
**Theoretical guarantee**: If the dataset is linearly separable, then the perceptron algorithm will always find a function that separates the data.- If data is not linearly separable, you can repeat step B of algorithm a fixed number of times.

This is my first time writing a tutorial. Hope you enjoyed, and I'd love some feedback. I tried to make the tutorial as clear and to the point as possible.

Read more…(318 words)

Mark as completed

Previous

What is Machine Learning? Why Machine Learning?

Next

Tutorial: Linearly separable data

Part of courses:

About the contributor:

Kavita RawatMachine Learning and Deep Learning enthusiast

100%

Loading…

Have a question? Ask here…

Post

Part of course:

Perceptron algorithm

- The perceptron
- The learning algorithm
- Why does the algorithm work?
- Algorithm at work:
- Notes:

Ready to join our community?

Sign up below to automatically get notified of new courses, get **reminders** to finish ones you subscribe to, and **bookmark** lessons to read later.

Continue with Facebook

— OR —

Your Full Name

Email address

I have an account. Log in instead

By signing up, you agree to our Terms and our Privacy Policy.

Popular Courses

New Courses

About Us

Get in touch

Copyright 2016-18, Compose Labs Inc. All rights reserved.