Skip to content

用Tensorflow2.0進行外匯收盤價預測, 使用LSTM, 1d-CNN, Seq2Seq

Notifications You must be signed in to change notification settings

purelyvivid/AIA_st2_kaggle_ml4

Repository files navigation

AIA_taiwan_southern_2nd

臺灣人工智慧學校(AIA)南部分校技術班第二期

Kaggle 競賽 - 透過training data預測最後30筆外匯收盤價

https://www.kaggle.com/c/aia-st2-forex-predict-overview

DL framework

Tensorflow 2.0

Data

~70000筆訓練資料 預測30筆接下來的收盤表現

Metric

MSE Error of 收盤價('Close')

Method & Result

  1. LSTM, mse_error: ~0.0006
  2. 1dCNN, mse_error: ~0.0010
  3. Seq2Seq, mse_error: ~0.0012

Remark

  • 簡單起見, 三種方法都只用了收盤價('Close'), 但不代表其他欄位沒有參考價值
  • 雖然LSTM錯誤率最低, 但不代表最適合用來預測股市, 只是此資料集剛好如此而已
  • Seq2Seq模型太過複雜, 不適合此任務, 只是寫來當作tf2練習

About

用Tensorflow2.0進行外匯收盤價預測, 使用LSTM, 1d-CNN, Seq2Seq

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published