#project structure
--./data: includes 2 files, one for train model and the other for test
--./model: model saved every 100 steps when training, can be used when test
--train.py: the train model program
--model.py: lstm-model file
--preprogress.py: to generate data for training and testing
--prediction.py: test file using model generated by train.py
27/07/2018 update
update 2 more models: single layer BiLSTM model and multi-layers BiLSTM model
the multi_bilstm model achieves the best effect in all 3 models(with same hyper-parameter setting)
01/08/2018 update
the earier version uses only close price to predict close price, in this version I increase input data to 4 dimentions, using [close, high, low, open] to predict future move of stock price.
with same train settings, this version achieves better effect than last one.