The Bucky model currently supports Linux and includes GPU support for accelerated modeling and processing.
GPU support is provided via CuPy. You’ll need to ensure that your system is compatible with CuPy’s requirements, namely that you have an NVIDIA CUDA GPU and CUDA Toolkit version 10.2+ installed prior to installing bucky.
Python version v3.8.0+ / v3.9.0+ / v3.10.0+. If your system has an older Python release we recommend installing bucky in an anaconda environment. Instructions to install conda are availible in the conda documentation.
git must be installed on you system and in you PATH.
This will install bucky for execution ONLY. If you plan to develop the model you’ll need to install it via the instuctions found here
Install bucky via pip:
pip install bucky-covid
Setting a working directory#
Bucky will produce multiple folders for downloaded historical data and outputs. It’s recommended to put these in their own directory, for example ~/bucky, and excute the bucky CLI from that directory.
BUCKY_DIR=~/bucky mkdir $BUCKY_DIR cd $BUCKY_DIR
The location of these directories can be globally specified in the configuration files. TODO link to config
Running the Model#
In order to illustrate how to run the model, this section contains the commands needed to run a small simulation. First, you have to download the input data required for a simulation:
bucky data sync
By default this data will be save to <pwd>/data.
You can now run the model, calculate quantile estimates and generate some plots. For example, to run the model with 100 Monte Carlo iterations and 20 days:
bucky run -n 100 -d 20
Equivalently, you can run each step on it’s own:
bucky run model -n 100 -d 20 bucky run postprocess bucky viz plot
Running the model will produce output csvs and plots located in the specified output_dir, by defualt this will be located at <pwd>/output.