Bullishness Indicator Vs BTC Price/Return
Recursive Neural Nets Prediction
Recursive Neural Nets Loss [Without BTC/USDJPY Log Ratio]
Recursive Neural Nets [Without BTC/USDJPY Log Ratio]
Regression Display: Covariance
SVM Classification Report
SVM Confusion Matrix
Augmented Dickey-Fuller Test between Bullishness Indicator & BTC Price/Return
SVM Learning Curve
SVM ROC/AUC Curves
I am driven by pursuing a career in trading and whatever I do is somehow related to this field. I am quite interested Bitcoin since August 2016 and I have been weary of its price movement during the 70M hack ever since. Across forums and also from friends who actually trade these cryptos in Prop Trading firms, it was a surprise to me that many end up with positive gains solely relying on technical analysis. I figured that since platforms such as Twitter and StockTwits are filled with these types of traders, it would be interesting to dig further and see for myself if there is actually value in the info that I am about to share with you.
What it does
I am providing you with my dataset. You can train ML algorithms to make them better at predicting live Bitcoin Tweets. On a daily basis, you will be able to generate a historical pseudo Bitcoin Bullishness indicator, which tells you how strong the market is feeling right now on Bitcoin prices. You can validate whether the indicator can be trusted or not by running my regression tests which contains regression, cointegration/unit root tests, and so on. I have also included a Recurrent Neural Network model to show you predictive results that I generated.
How I built it
Tensorflow, Scikit-Learn as my ML sources, Twitter API/Thomson Reuters Trkd API/Bitstamp API to obtain financial time series, which are stored into MySQLdb, Statsmodels in Python to generate my statistical tests.
Challenges I ran into
This is the first time I tried using Tensorflow, so the learning curve is quite steep. The Bitcoin Tweets that I gathered were very hard to work with, it was extremely time consuming to produce 1000 rows of BTC tweets.
Accomplishments that I'm proud of
Learned about the basics of Recursive Neural Nets through reading several papers. Executed my idea from step A to Z. The next step is very clear to me especially in terms of knowing what types of data to gather, and how to make the entire program run faster...
What I learned
Tensorflow, Scikit-Learn, writing a wrapper class for Thomson Reuters Trkd API.
What's next for TheBitcoinSentimentDataAnalysisV2
Optimize my hyperparameters for RNN, make the process more scalable Looking to improve my understanding of NLP techniques, hopefully someday making this project into a separate module in my algo project ;)
Thank you very much for taking the time to read!
Sincerely, Terence Liu