Deeprecip is a recurrent neuronal network with an architecture tailured to produces short term precipitation prediction. Images from weather radars serve as inputs.
The network is implement in Chainer, a deep learning framework based on the “Define-by-Run” scheme.
This is work in progress. See this slides for more information.
- Install required system packages (instructions for Ubuntu)
$ apt install gcc-4.8
$ apt install python3-dev
$ apt install libhdf5-dev
$ apt install libfreetype6-dev
$ apt install graphviz
$ apt install python3-mpltoolkits.basemap
Note, currently a pip install of basemap seems not possible.
-
Install CUDA and CuNN. Make sure that you run
gcc
andg++ on
version 4.8! Useupdate-alternatives --config
to switch versions. -
Clone package
git clone https://gitlab.com/scheidan/deeprecip.git
- Create a virtual environment and activate it
pyvenv venv
source venv/bin/activate
Note pyenv
requries python 3.
- Download the required python packages
(venv) $ pip install -e .
The -e
flag means "editable" and creates links to the python files of
the package (instead of copying them). If you update the package, the
new files become available immediately. (Thanks Uwe!)
See run.py
for an example.