Reproducing prospective test results for RELM and hybrid earthquake forecasting models for California
This repository provides data and code to reproduce prospective test results for six RELM and sixteen multiplicative hybrid earthquake forecasting models for California, reported by Bayona et al. (2022). This experiment takes about 2 hours on a modern desktop computer if the number of simulations per forecast and per test (expect for the Poisson and NBD N-tests) is set to 1000.
The scripts to execute the experiment can be found in the code directory of this repository. This folder contains two
.py files, namely download_data.py and reproducibility_hybrids.py, which download forecast files from
Zenodo, run the computations, and create and store the figures presented in the manuscript. Finally, the
run_all.sh file, in the top-level directory, is a shell script that runs the entire experiment by simply typing bash ./run_all.sh
in the Terminal.
python=3.10.1
numpy=1.21.5
pycsep=0.5.0
In order to run this reproducibility package, the user must have a pycsep environment installed and running on her/his machine ('gji-hybrids' in this example). The easiest way to install
pycsep is using conda, however; it can also be installed using pip or built from source (see the Documentation on how to install pyCSEP).
conda create -n gji-hybrids
conda activate gji-hybrids
conda install --channel conda-forge numpy=1.21.5 pycsep=0.5.0
In addition, the user must have access to a Unix shell that has python3 installed with the requests library (it should be provided with pycsep=0.5.0). If not, she/he can install this library using:
conda install requests
These instructions assume that the user is "within" the environment, with python3 and the requests library installed. Thus, running the experiment is as simple as:
git clone https://github.com/bayonato89/Reproducibility-hybrids.git
cd Reproducibility-hybrids
bash ./run_all.sh