spidoche/rnn_stock_predictions

Stock predictions with RNN

rnn_stock_predictions

data Crawling, Pretreatment, Processing, Training, Model Visualization -> AUTOMATION

requirments

• Python 3.5.3
• tensorflow 1.1.0
• Pandas_datareader
• numpy
• matplotlib
• datetime

Run

• python apple.py
• python kospi.py
• tensorboard --logdir=./tensorflowlog
• tensorboard http://localhost:6006/
  graphs

Model

• RNN + Fully connected layer
• Train : Test = 70 : 30
• Train Period = 2010.1.2~2017.5.27
• Predictions Period = 2017.5.28~2017.6.7
• Real Period = 2017.6.8

Results

• Alphabet 0.050(RMSE) 1004.28(Real) 1001.59(Predictions)
• apple 0.020(RMSE) 154.99(Real) 155.37 (Predictions)
• berkshire 0.016(RMSE) 250305(Real) 249621 (Predictions)
• hyundai_motor 0.020(RMSE) 160000(Real) 159000 (Predictions)
• kospi 0.022(RMSE) 2363.57(Real) 2360.14(Predictions)
• samsung_electronics 0.022(RMSE) 2258000(Real) 2265000(Predictions)
• sk_hynix 0.024(RMSE) 56700(Real) 56500(Predictions)
  
• berkshire
  

Reference

• https://github.com/hunkim/DeepLearningZeroToAll