To train the neural network you use historical data ?
We have a training sample is collected on the basis of technical analysis.
do you develop your own machine learning algorithm or use well known? and how long does it take to train with refreshed set of historical data?
Yes, we develop our own algorithm on the basis of the LSTM network and gradient boosting.
Historical data are not used!!
We manually collected the sample on the basis of technical analysis patterns.
We complemented signals are MACD.
Next year we will launch a fundamental analysis.
News, tweets, posts will be analyzed by the algorithm of the tone of the text and of the received signal will be used in the algorithm technical analysis.
Also, we'll extend the algorithm of the technical analysis will be added to the Elliott Wave.
Training takes place on a powerful GPU Titan Xp, and takes approximately 40 minutes.
We use the library Keras+Theano for gradient boosting is used XGBoost.