For those of you who are interested in machine learning, particularly with deep learning, but have no idea where to start, I just wrote a new post on my blog explaining some practical basics on machine learning and how to get started on your own computer.
I show how anyone can get their hands dirty and get started with building neural networks using Keras and Tensorflow libraries. At the end, we will train a basic convolutional neural network that can recognize handwritten numbers. The framework that I outline is very similar to the one Google used to design AlphaGo (Google Deep Mind are the developers of Tensorflow), which just recently beat the #1 professional player at the game of Go.
As the blog post is quite long, with several pictures, I will include the link here (I hope that is okay moderators):
I would love to hear what you guys think and if you have any feedback. If there is enough interest, I would definitely love to write more about the subject (as this is my primary method of trading) so any suggestions on topics you’d like to read more about would be awesome!
I have experimented with ML techniques before in the past and found that trying to build a model for price forecasting was just a waste of time. At that time, deep nets (RNNs in particular) were still difficult to train, so it was mostly simple neural networks. Instead, I use quite a few ML models to manage my portfolio and adjust exposure to the market, and still do with good success.
To be honest, I am still learning as I go with Keras and TF. I think deep learning has a lot of potential, especially with LSTM models solving the vanishing gradient problems we used to have in normal RNN models. In particular, I’d like to take a look at price/regime forecasting using LSTM RNN models. Who knows, maybe it might have better results.
