heatherjia / deepmhd Goto Github PK

These notebooks are provided to reproduce all data-containing figures and all results from

Do Androids Dream of Magnetic Fields? Using Neural Networks to Interpret the Turbulent Interstellar Medium

by J. E. G. Peek and Blakesley Burkhart, accepted ApJL, https://arxiv.org/abs/1905.00918

With these notebooks, and the associated data here, you should be able to

• extract normal and Fixed Fourier Power (FFP) images from the turbulent simulations and save them as files for training and test data
• train networks to classify these images
• evaluate the networks (either your own trained networks, or the ones provided by us at here)
• run the saliency map analysis
• make figures 1 and 3 (without annotation)

These notebooks have a few dependencies:

• Python 3
• numpy 1.14.5
• tensorflow 1.6
• keras 2.2.4
• sci-kit-image 0.13
• matplotlib 2.1.2
• tqdm

There should not be strong hardware requirements to make figures and evaluate networks. To train the networks we used a NVIDIA Tesla P100 GPU, and we do suggest GPU acceleration of some kind for timely network training.