Inspiration
This project -which was developed during CopenHacks 2017- is a very simple demonstration of a stock market predictor which uses TensorFlow. Successfully predicting the future value of stock markets can surely make one a billionaire, but as much times as it has been unsuccessfully attempted by bright mathematicians working for big Wall Street enterprises, it is mostly thought as impossible and absurd.
Anyways, it seemed to me a fun idea to work on something related to economics and especially related to deep learning. I wanted to have a first grasp on deep learning and this has been the opportunity to do so in a challenging and fun environment, at my own pace. I have had a great time coding this.
What it does
The predictor trains a very simple quadratic function by correlating historic data from a specialised economic news feed (which is analysed using Microsoft Cognitive Services to detect the sentiment which it transmits) and historic data from the stock market. From here, with large databases and a well-trained model, interesting prediction results could be produced by monitoring live the economy news feed.
How I built it
This project was built using Python on Jupyter Notebook, and also the TensorFlow library -especifically designed by Google for Deep Learning purposes.
The Microsoft Cognitive Services API (the Text Analysis one) was used to detect the sentiment enclosed in the news, which was directly correlated by the model with the fluctuation in the stock market.
Challenges I ran into
One of the biggest challenges that I ran into was making Microsoft Cognitive Service's API work -it took me a while until the API accepted my input data as valid. Then, the biggest problem of the project, which has an enormous impact on it, is the lack of good, reliable API providing stock market information.
A free API -Markitondemand- providing a bit of information on them, but clearly insufficient for a professional project. The data sets obtained from it are very small, and thus, the results of this project are limited and circumscribed to this fact.
Accomplishments that I'm proud of
I am proud of having been able to develop this idea from the beginning to the end by myself, without needing much help. It is the first time that I hack alone, and though I reckon that it is more fun and less stressful to hack in group, this experience has allowed me to better know how I work by myself and thus improved my self organization skills.
What I learned
I learned to use TensorFlow, which I had never used before, and very basic concepts related to deep learning. I have refreshed the knowledge that I had on Python, which is also very cool and useful.
What's next for Feel The Market
Feel The Market has a great potential if a big database is provided for it. In that sense, in the future, the sources of information should be diversified and bigger ones should be used. This would provide more confidence in the obtained output results.
Log in or sign up for Devpost to join the conversation.