While there are a large number of useful tutorials that explain machine learning to coders, most seem to lack a basis in curriculum development. By that, I mean that they often show one how to do intermediate or advanced machine learning without covering the basics. This may be ideal for computer science graduate students who have some experience in machine learning, but it is not particularly helpful for someone in psychology interested in using these techniques from a very different perspective.
I suppose what I am proposing is a long series of posts that, when combined, would constitute a course or series of courses in machine learning and its applications from the very beginning. Each decision would be explained by describing what happens when we do the alternative. Why do we choose this learning rate? What happens if it is larger? What happens if it is smaller? What are the alternatives to a constant learning rate? Each post could explore one such topic or one specific application, starting with the smallest of models and building toward the state of the art.
Perhaps, in some small way, this more structured educational approach will help machine learning break out of the computer science and technology sphere.