Detecting real hand-written digits with machine learning

Matplotlib is my go-to tool for plotting in Python; I like that it is essentially infinitely customizable, allowing you to create polished looking plots. But, it suffers at least a few drawbacks that may make it undesirable for some applications:

In one of my research projects we pass particles through small microfluidic channels, and on occasion after an experiment I don’t clean up as well as I should. Here are some chips I found lying around that I thought would be cool to image. The small spheres are polystyrene beads, and the solution/precipitate is KCl. For reference, the beads are 10 microns in diameter, or around a tenth of the thickness of a strand of human hair.

In this post I’ll write about my attempt at the digit recognition Kaggle competition. The goal is to accurately recognize single hand-written digits, which are provided as two-dimensional grayscale images. Each pixel element is an integer value in the range [0, 255].