Team:
- Animesh Mishra
- Konstantinos Metallinos
- Maximilian Hoefl
Dataset:
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:
- Get the dataset
- Extract the
GEFCom2014 Data/Load/Task 1/L1-train.csv
file from archive - Put it in the
/data
folder - 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 scriptsdata
: has original data as well as data generated during feature processingmodel
: has parameters from model (e.g. during pretraining or final modeling)output
: has output figures