time-series-forecasting-keras

The experimental source code of Paper: Time Series Forecasting using GRU Neural Network with Multi-lag after Decomposition, ICONIP 2017. paper, HomePage

Requirements

• python 3.6.3 (Anaconda)
• keras 2.1.2
• tensorflow-gpu 1.13.1
• sklearn 0.19.1
• numpy 1.15.4
• pandas 0.23.4
• statsmodels 0.8.0
• matplotlib 2.1.0

Model

• LSTM
• GRU
• RNN
• MLP
• SVR
• ARIMA
• time series decomposition

Code

• data
• models
  • decompose.py: time series decomposition
  • MLP.py: MLP network
  • RNNS.py: RNN faimily network, including RNN, LSTM, GRU
  • SVR.py: SVR model
• naive_MLP_forecasting.py
• naive_RNN_forecasting.py
• naive_SVR_forecasting.py
• decompose_MLP_forecasting.py
• decompose_RNN_forecasting.py
• decompose_SVR_forecasting.py
• ARIMA.py (to do)
• util.py: load data, pre-processe time series, including multi-lag sampling
• eval.py: calculate the metrics
• subseries_plot.py: plot figure of time series decomposition