StockRecommendSystem

Main Requirement:

Python 3.5.2
Keras 2.0.6 TensorFlow 1.2
pymongo
tqdm nltk
googletrans

Install

brew install mongodb --with-openssl
brew services start mongodb
mongod --dbpath (Your Porject Folder)/Data/DB

When you storing stock data with mongodb mode, you may meet too many open files problem, try the following codes in command line:

sysctl -w kern.maxfiles=20480 (or whatever number you choose)
sysctl -w kern.maxfilesperproc=18000 (or whatever number you choose)
launchctl limit maxfiles 1000000 (or whatever number you choose)
brew services restart mongodb
mongodump -h localhost:27017 -d DB_STOCK -o ./


Data Fetching:

Cover stock related data fetching, storaging in either MongoDB or CSV mode (See config.ini [Setting] sector for more detail).

  1. Stock:（NSDQ, NYSE）-> US, (HKSE) -> HK, (SSE，SZSE) -> CHN
  2. Earning: US stock market earning info.
  3. Short: US stock market short squeeze info. (Require Multi IP Routing Support)
  4. News: NewsRiver
  5. Media: Twitter Data

Data Structure

** US Stock List **   
DB   : DB_STOCK   
SHEET: SHEET_US_DAILY_LIST   
ITEM : symbol, name, market_cap, sector, industry, stock_update, news_update   

** US Stock Daily **   
DB   : DB_STOCK   
SHEET: SHEET_US_DAILY_DATA   
ITEM : symbol (stock symbol)   
       data -> [{date, open, high, low, close, adj_close, volume}]   

** US Stock Earning **   
DB   : DB_STOCK   
SHEET: SHEET_US_EARN   
ITEM : symbol （date）   
       data -> [{date, symbol, analyist, estimate, actual, surprise}]   

** US News **   
DB   : DB_STOCK   
SHEET: SHEET_US_NEWS   
ITEM : symbol, date, time, title, source, ranking, sentiment, uri, url, body_html, body_eng, body_chn

Run

cd Source/FetchData
python Fetch_Data_Stock_US_Daily.py

Stock Prediction:

Under Development...


Stock Processing:

Correlation

   Company1 Company2  Correlation  
       QQQ     TQQQ        0.999
       IBB      BIB        0.999
      INSE     XBKS        0.999
       JAG      JPT        0.999
      ACWX     VXUS        0.995
      IXUS     ACWX        0.993
      VONE      SPY        0.992
      IXUS     VXUS        0.991
      VTWO     VTWV        0.988
       NTB      FBK        0.988
      GOOG    GOOGL        0.987

Run

cd Source/StockProcessing
python Correlation_Stock_US.py


Reinforcement Learning:

This sector is directly clone from: Link

More in mind:

  1. The approach use only "Adj Close" price as input, it's supposed more features combinations shall be joined to the party.
  2. The Trading Strategy is a little mediocre and limited, better rewrite it.
  3. At most only two tickers are allowed in the trading system, rewrite it.

testing output:

init cash: 100000
Columns: [AMD, NVDA, SPY, ^VIX]
Index: []
Runner: Taking action 2016-03-16 00:00:00 buy
Runner: Taking action 2016-03-17 00:00:00 buy
Runner: Taking action 2016-03-18 00:00:00 hold
......
Runner: Taking action 2017-06-12 00:00:00 buy
Runner: Taking action 2017-06-13 00:00:00 buy
Runner: Taking action 2017-06-14 00:00:00 buy
Final outcome: 121500.348294

Run

cd Source/ReinforcementLearning
python runner.py


ToDo:

More AI approach will be arranged and upload ASAP