Sunspot Number Prediction

We use LSTMs to predict the sunspot number data from the observed data that is available (1818-2019).

Instructions

In order to use this code to predict sunspots:

1. Clone this repo.
2. Create the directories img/, models/ and logs.
3. Run the Jupyter notebook bin/main.ipynb.
4. Change the %cd to match the path to the head directory on your system.
5. Run all the cells in the notebook.

Requirements

• Python 3 (tested with 3.7.7)
• libs: numpy scipy pandas matplotlib tensorflow_gpu keras_gpu jupyter

Notes

• We have used Adadelta optimizer. You can use the Adam optimizer instead as well. Just change the trainer function in src/network.py. Both converge to the same overall minima, Adadelta converges from the top (underfits) while Adam converges from the bottom (overfits).