pred_finance

stock price prediction using machine learning

stock_DQN.py

• Deep Q-network implementation
• DQN model type include CNN and LSTM
• Target-freezing and DDQN are also implemented
• python stock_DQN.py for training, which will automatically search for pre-trained parameters to load
• python stock_DQN.py -m test for inference, which will show total reward and accuracy
• While infering, an action to close price figure will also be generated. Color red stand for correct action, color green stand for wrong action
• several training arguments can be modified inside the python script

stock_DQN_prioritized.py

• implement DQN with prioritized experiency replay. Reference
• Only CNN network are used in the current script
• Usage and features are same as stock_DQN.py

stock_supervised.py

• Supervised Deep Learning implementation
• CNN and LSTM are included
• Usage and features are same as stock_DQN.py

save_stock.py

• Saving stock data from yahoo_finance
• Calculating several index
• Currently, Index includes: High Low Open Close d_Close RSI9 RSI15 MA5 MA20 MA60 d_CO d_HL Adj_Close Volume VA/D d_VA/D %R8 %R21 DIF DEM d_MA5_20
• Before running th script, create a folder stock_data, where csv file will be stored
• Usage: python save_stock.py NAME INDEX [-h] [-e END_YEAR] [-s START_YEAR] [-r RATIO]
• NAME: the name of csv file to store
• INDEX: stock index of the company
• --start: the data will start from START_YEAR
• --end: the data will end with END_YEAR
• --ratio: when apply the ratio argument, data will be split into train and test csv file, which NAME_train.csv contains RATIO percent of data
• see python save_stock.py for more info

draw_stock.py

• draw the stock data from applied csv file
• Usage: python draw_stock.py FILE [-h]
• see python draw_stock.py -h for more info

stock_state.py

• implement class Stock_state
• can be used to generate next state for given action
• can be used to generate random-shuffle data batch for supervised learning
• can be used to generate none-random-shuffle test batch for supervised learning
• can also perform similar function as draw_stock.py
• usage: python stock_state.py COMP_NAME, e.g python stock_state.py TSMC