A series of iPython notebooks applying ML to recognizing MNIST data
This repository is my first on GitHub, and is meant to keep track of some of my earliest attempts at learning data science. It involves applying various machine learning approaches to a classic toy problem: recognizing handwritten numerals that have been curated by the MNIST. Specifically, I have accessed these data through Kaggle, and will post the notebooks and predictions to that site.
I preceded the exercises contained here with a reading of an ebook by Joel Grus meant to introduce neophytes to data science, which I found a useful without fully understanding all of its content. The digit recognizer is my first hands-on experience with some of these concepts, using a very convenient, sanitized data set.
My first attempt was with the random forest approach; then I proceeded to try a simple neural network, a combination of dimensional reduction and support vector machines, and intend to finish with a convoluted neural network. My choice of approaches is largely inspired by seeing the work others have done with this same data set on Kaggle.
I'm very new to Python, machine learning, and data science, and expect to make plenty of mistakes and awkward choices. Note, however, that my interest is mainly in experience as a data scientist, not a developer/programmer, so issues with the readability or efficiency of scripts is less important to me than questions regarding the applicability of approaches chosen, or drawing proper conclusions from the data. Beyond that, any feedback is always appreciated.