Introduction

A number recognizer application with almost95% accuracy written by Scikit-learn (sklearn) library as core of processes, pygame module used for draw app and Pillow library for editing images.

How to use?

I) First of all, while you are running the code, you have to be online.

II) Check you have this library in your enviroment:

1. sklearn
2. numpy
3. PIL
4. pygame
5. scipy(usually by installing sklearn you installed numpy and scipy by default)

III) Then run main.py

After taking 6 or 7 minutes (depends on your system), you will see small window (280 × 280) and you can draw a number between 0 to 9. And by pressing Enter you will send your image to program and it return it's guess and it continues till you close program by close button.

How program works?

First program gets MNIST dataset from internet and after AUGMENTING DATA (shifting image into left, right, down and up), program add them to our training dataset. Program starts to find best hyperparameters for KNN CLASSIFIER by GRIDSEARCHCV. Then it makes new class for knn classifier and pass best parameter into constructor of knn classifier. Afterward program run draw app. and gets images from user. It resizes image to 28 * 28 and converts images to numpy array to pass it into knn_clf.predict() function and it returns the anwser.