Open doncat99 opened 7 years ago
So far, the split
works for a single feature alone. I'll be extending this to multi-dimensions very soon. I'll try having it out at the earliest. Mind you this project is still at its very nascent stage.
https://github.com/christsaizyt/US-Stock-Market-Prediction-by-LSTM
the link above introduce average price "MA" and volume into the LSTM network, by presenting two different implementations (regression and classification) to predict stock market price. Hope it helps.
From my point of view, the two implementations shouldn't be isolated. The classification approach may be the based foundation, regression approach should based on the output classification approach provided. For more, DQN network could integrate on the upper side to supervise the whole learning process.
you can reference the gits below for more information. https://github.com/carpedm20/text-based-game-rl-tensorflow
Moving Averages have now been introduced. I still need to figure out how it could be embedded within the Share
object.
Classification-Regression seems to be a great idea! The reference seems great. I'll try having something on the lines of scikit-learn
's Class-Reg
I found that majority implementations of stock prediction toke only one parameter as input, either 'Open' or 'Close' price. While I wonder if it is possible to combine 'Open' price and 'Volume' value together as input to the network to learn the relationship in between?