Skip to content

Latest commit

 

History

History
77 lines (57 loc) · 2.78 KB

README.md

File metadata and controls

77 lines (57 loc) · 2.78 KB
Raw

AutoTradingSystem

키움증권 자동매매프로그램

Basic

  1. Trader
  • tf_gpu.py : main file for Reinforcement learning
  • agent.py : decide action to trade and validate
  • policy_learner.py : define Reinforcement Models
  • policy_network.py : define Neural Network Models
  • visualizer.py : visualize chart data to graph
  1. Bridge Server For Win32
  • kiwwom_bridge_flask.py : bridge server connecting between Trader to Kiwoom OpenApi+
  1. DataBase
  • store and share database with sqlite3

Requirements and Environment

Use Anaconda (Python distribution platform)

  1. Trader
  • python 3.6
  • trader is using tensorflow, numpy, pandas, scipy, talib
  • pip install TA_Lib-0.4.19-cp36-cp36m-win_amd64.whl tensorflow-gpu==1.15.2 pandas scipy requests matplotlib mplfinance keras==2.2.4 h5py==2.10.0
  1. Bridge Server

Ideas

  1. kiwoom OpenApi+ is available only in 32bit environment
  2. Seperate Kiwoom trader and Actor(Reinforcement learning Agent)
  3. Need Simple Api Server to connect between open api and deep learning

Details

  1. Trader
  • Trader is learning stocks data to predict when is time to buy,sell or hold.
  • Trader decide action using A2C Reinforcement Models and LSTM Neural Network.
  • Trader is getting reward under conditions which are 60% win/loss
  • Stocks data contain OHLC Daily chart data, Kospi Daily char data, MACD histogram, RSI
  1. Bridge Server
  • To get and update stocks data, use Python Flask for Api Server
  • To use kiwoom OpenApi+, import koapy library
  • The Bridge Server is log in kiwoom server and get stocks data from kiwoom with koapy
  1. Database
  • Use sqlite3 to store data
  • The Trader is getting chart/kospi data and learning from DataBase/DB/stocks.db
  • The Bridge Server is storing chart/kospi data to DataBase/DB/stocks.db from kiwoom server

Installation

  1. Install Anaconda
  2. Make python3.7 32bit and python3.6 in venv (using conda create)
  3. Install each python packages with requirements
  4. Install CUDA Toolkit (https://developer.nvidia.com/cuda-toolkit-archive) for GPU

Testing

  1. To learning

    • activate py36
    • python Analysis\tf_gpu.py --stock_code {$} --rl_method {$} --net {$} --num_steps {$} --output_name {$} --learning --num_epoches {$} --lr {$} --start_epsilon {$} --discount_factor {$} --ver {$} --delayed_reward_threshold {$} --start_date {$} --end_date {$} --value_network_name {$} --policy_network_name {$}

  2. To Trading

    • activate py37_32

    • python Bridge\kiwoom_bridge_flask.py

    • activate py36

    • python Analysis\tf_gpu.py --stock_code {$} --reuse_models --rl_method {$} --net {$} --ver {$}