I want to start this post with a little insight into why I chose the title. Machine learning and Artificial Intelligence has been a subject that has both interested me and terrified me. So naturally I decided to do a course long project on a concept that I knew nothing about. I came into this assuming that I was in for a long term where I would be spending 30 hours a week trying to wrap my head around how AI works and how to code something that learns. But after a week in and some preliminary research, I can say with confidence that Machine learning and AI isn’t nearly as terrifying as I thought, hence the myth. Now onto our project of AI/ML Bot for Trading Bitcoin.
I’ve done about 10-15 hours of pure reading and watching videos, coupled with a couple of group meetings with our highly adept project mentor, thanks Chester, and I think I am well on my way to wrapping my head around what needs to be accomplished. So long 30+ hour weeks, I’ll see you in hell. But I digress, we have landed on Python and TensorFlow as being our main coding language/technology.
Python was chosen due to the fact that our project mentor specified it as being an excellent choice. From the reading I’ve done on Machine Learning, it seems that TensorFlow (a Python Library) is the go to library for training a model how to perform a task. So that, coupled with Chester’s recommendation made it a no-brainer for what technology to use. We will also be using Quant Connect for the actual trading aspect of the project. This was chosen because, once again, it is the go to API for algorithmic trading and integrates well with TensorFlow. Quant Connect will provide us with datasets for training and a way to connect with a trading platform such Binance to perform actual trades.
Let’s start off with TensorFlow but before I dive in, I want to remind you that I’ve only just begun to research and try to understand the technologies we will be using. From my understanding, TensorFlow will be used to actually train our model. We will input Features into the model (historical data) and expect the model to perform a certain task under given conditions. For example if the current moving average is greater than the 200 day moving average than we want to Buy (rough example, don’t laugh). The main concepts we want to train our model through use of TensorFlow will be a Buy strategy, a Sell strategy and a hold strategy. We will be using 3-4 different features with 4-5 datasets for each feature to actually train the model.
Once the training is completed and we have a decent model that performs comparable trades to what we are expecting than the fun begins of performing actual trades! Now, I haven’t scratched the surface on Quant Connect yet, however this is where Quant Connect comes in. This is where we will integrate our model and use actual data to perform trades. Eventually we will connect with Binance API to perform real, hopefully profitable trades.
And that my friends is all I have to say at this point. There is a long journey ahead and I hope to come out a millionaire through use of a profitable Bitcoin trading bot. Stay with me for the journey and hopefully you can learn something too.
Signing Out,
Geoff Miller
Leave a Reply