Team:

• Animesh Mishra
• Konstantinos Metallinos
• Maximilian Hoefl

Dataset:

• Global Energy Forecasting Competition 2014 (details)
• Download link.

Software dependencies:

• Python 2.7.x and libraries (numpy, pandas, scikit-learn, matplotlib, keras, theano)
  • Version compatibility: Keras 1.0.0 with Theano 0.8.0.dev0 as backend

Usage:

1. Get the dataset
2. Extract the GEFCom2014 Data/Load/Task 1/L1-train.csv file from archive
3. Put it in the /data folder
4. Go to /code folder. Run the python/R scripts in the following order:

  dataVisualisation.py -> for visualizing the data
  Baseline_model.R     -> implements a baseline model (ARIMA) in R
  getFeatures.py       -> extracts features for python script
  mlpPretraining.py    -> pretraining the MLPs. Includes station selection.
  mlpEnsemble.py       -> trains the ensemble of MLPs
  makePredictions.py   -> predictions and plots
  

Folder description:

• code: has python/R scripts
• data: has original data as well as data generated during feature processing
• model: has parameters from model (e.g. during pretraining or final modeling)
• output: has output figures