Coinseer

Inspiration

The ghost of Bernie Madoff's career came to me in a dream and told me I could get super rich by predicting bitcoin prices.

What it does

Given properly formatted and aligned data, can detect any effect of changing financial markets on the price of bitcoin, predicting whether it will move up or down. No idea if that's what it actually does.

How I built it

The original plan was for other members of the team to write a neural network library of our own using OpenCL and Visual C++, while I used node.js to pull down as much relevant data as possible and send it by websocket to a C++ server, where it could be parsed and fed into the maw of the network. Bizarrely, this ran into problems and was scrapped. And then everyone went home to nap and the only person who got back before the library locked down was me. Now I'm using a couple of python scripts to query GDAX and Google Finance for data, format them, and store them in csv files. Another pulls from these files and initializes a super basic neural net from the scikit-learn library with the data, and then the net trains itself.

Challenges I ran into

Using Winsock to set up a C++ server. Learning node. Learning node again, except for Visual Studio. Parsing vast reams of financial data. Trying to get the vast reams of financial data properly time aligned with one another. GDAX rate limits public client requests to 3 per second and historical pricing data to 200 items per request, so I had to set up a script to get an entire year's worth of data in pieces.

Accomplishments that I'm proud of

Managed to learn and discard a language(node.js) in a single sprint. Pulled enough data to (hopefully) build and train a neural network with the ability to correctly predict bitcoin prices.

What I learned

A good bit of node, a bunch of python libraries, GDAX API, enough about neural nets to know I've only just scratched the surface of a galaxy-sized topic.

What's next for Coinseer

Accounting irregularities.

Built With

• python
• scikit
• scikit-learn