nina-om/swirlnet

Deep learning tutorial for predicting low-level rotation.

Swirl Net

This is a tutorial on using deep neural networks to predict intense rotation near the surface in thunderstorms.

Requirements

• jupyter
• numpy
• scipy
• matplotlib
• xarray
• pandas
• tensorflow
• keras

If you want to install these libraries on your local machine, the Miniconda Python package manager is recommended. Please follow the link and instructions to install the appropriate Miniconda for your OS. After Miniconda is installed, either install the packages directly with conda install or create a conda environment (bash shell required). Tensorflow with GPU support requires a NVIDIA GPU with CUDA and cuDNN to be installed. Specific installation instructions are beyond the scope of this tutorial, but more information about installing tensorflow can be found here. The easiest way to get access to a GPU with deep learning capabilities is to use a Deep Learning image on AWS, Google Cloud, or Azure or find an appropriate docker container. To install the necessary libraries without GPU support, please use the following commands:

conda create --name deep -c conda-forge python=3.6 numpy scipy matplotlib pandas netcdf4 xarray jupyter
# Activate the environment
source activate deep
# Install tensorflow and keras from their PyPI binaries
pip install tensorflow
pip install keras