Perceptron for digit recognition

This project uses a perceptron network for digits recognition. It uses the data in training_data.csv to train each perceptron and then set ups a web server to listen for POST /recognize requests at port :8000.

Because single layer perceptrons are binary classifiers, the script creates and trains one perceptron for each class that needs to be recognized at startup. When it needs to recognize a character, it sends the data to each perceptron and returns the one that gets fired.

You can easily edit the data in training_data.csv in Excel or Numbers to add or change the recognized digits.

An example of the data contained in the training set:

Uno	0	1	1	0	0
	0	0	1	0	0
	0	0	1	0	0
	0	0	1	0	0
	0	0	1	0	0
	0	0	1	0	0
	1	1	1	1	1

• The first column of a class is the class name or tag
• The rest of the info is a matrix of 5x7 representing the character to be recognized
• The script supports char matrices of any size but all characters must be of the same size

Requirements

This project depends on:

• Tornado

Running

Run with:

python perceptron.py

And open localhost:8000 with your browser

Credits

Basic Training and Dot Product functions based on example found in Wikipedia's Perceptron Article