I have just started doing machine learning using Keras and Tensorflow on Jupyter Notebooks / Anaconda.
Digit Recognition is my first machine learning project. It uses the MNIST database of labelled handwritten digits (about 70,000 of them) to find an algorithm to work out your digit. The digits are stored in single dimensional arrays of opacity. The biggest difficulty with the project was exporting the trained model from a Jupyter Notebook to Unity. The predictions aren't as good as I wanted because I couldn't export my Convnet model (which looks at shapes wherever they are in the image) and so I had to use a Dense network (which looks at pixels).
The first six digits in the MNIST data base.
You click and drag to draw a digit. You can draw anywhere in the second blue box but I recommend you draw in the darkest box. When you are done press recognise.