Citing: If you use this framework please cite as follows:

@inproceedings{2017AGUFMSM23A2591C,
    author = {{Cheung}, C.~M.~M. and {Handmer}, C. and {Kosar}, B. and {Gerules}, G. and 
{Poduval}, B. and {Mackintosh}, G. and {Munoz-Jaramillo}, A. and 
{Bobra}, M. and {Hernandez}, T. and {McGranaghan}, R.~M.},
    booktitle = {AGU Fall Meeting},
    title = {Modeling Geomagnetic Variations using a Machine Learning Framework},
    address = {Long Beach},
    year = {2017}
}

This was tested on P100 nVidia GPU and nVidia GTX 1060 GPU. Should work if enough RAM on computer and fast enough GPU.

python, anaconda, keras-gp, scikit-learn, pandas

For some of the csv files c++ programs were used for conversion. They can be found in cpp_files subdirectory. These are placeholder programs and probably should be converted to python for c++ phobic people. Check the README in that subdirectory for further details.

For either Project 1 or Project 2 edit the config.cfg file to point to the appropriate directory and import data.

Project 1: geomag with omni solar wind data Two files to run to explore LSTM with geo magnetic and solar wind data. Each one is a different take on LSTMs.

python LSTMarrayprediction.py

or

python lstm_multi_channel.py

Project 2: geomag, omni solar wind data and kp index

python kp_regress.py

Contact Information:

Mark Cheung cheung@lmsal.com

George Gerules ggerules@gmail.com or gwgkt2@mail.umsl.edu