There is a ton of data out there on the stock market. However, sorting through all the data for meaning can be difficult without a helping hand. With Gofor, you can make educated decisions with stocks through machine learning.
What it does
Gofor is a webservice and SMS chatbot that offers advice and predictions for the stock market. Powered by machine learning, NLP, and sentiment analysis, Gofor is a powerful tool that can be used by every trader. The webservice (www.goforanalytics.com) shows daily predictions for low, high, and closing prices of the next trading day. On top of that, a single click on a company will bring relevant news articles before you. If you are unable to secure an Internet connection, the Gofor SMS chatbot can answer queries for desired stock information.
How we built it
We used opensource stock data from IEXTrading to train a Keras model in Tensorflow. Available datasets include daily low, high, open, close prices, volume, news articles and more for every company on the stock market. Our model took past 100 days of low, high, and close prices and was trained to predict low, high, and close prices for a given future day. We migrated this model as well as several supporting python scripts to Google cloud, where they are queried by another python script that supports our apache server on Linode.
Challenges we ran into
The APIs we used were well documented and provided a great deal of functionality, so most of our challenge did not come from implementing specific components, but rather stitching everything together. We had to write 3 python flasks, have 2 different JSON parsers, and figure out how to serve dynamic data on a webpage.
Accomplishments that we're proud of
This is our first time using machine learning, we are pleased with how the model turned out. It makes some pretty spot on predictions, and because we were able to export it to google cloud, we can harness it in a variety of applications.
What we learned
Collectively we learned a lot of skills! We divided the components to make production more efficient, and so we each had our own challenges.The top thing each of us learned: "how to host a python script on google cloud," "how to transport data with JSON files," "how to identify good data for a Machine Learning model," and "why API keys should not be uploaded to git."