All files needed to build this image are available at AEADataEditor/docker-aer-2023-0700.
This Docker image is meant to isolate and stabilize the primary environment used for the analysis in AER-2023-0700. It does not cover all software packages (MATLAB is missing), and it is not optimized for compute speed, only for portability.
- Docker
- Access to Docker Hub
- Gurobi license (see https://license.gurobi.com/manager/doc/overview for details on how to obtain a license)
- We used a "Temporary Academic" license for container/machine deployment. The license is limited to 2 simultaneous sessions, but can expand up to 5 sessions for up to 5 minutes (this is a limitation for part of the analysis)
- Mosek license (see https://www.mosek.com/products/academic-licenses/). We requested a "personal academic license".
- Stata license
- If you already own a copy of Stata, you can use the
stata.licfile from your installation.
- If you already own a copy of Stata, you can use the
- Some of the scripts in the replication package require a job scheduler to run up to 100 parallel jobs. The authors and we used SLURM.
Copy all three license files into the current directory. The run_docker.sh script will look for the licenses in whatever directory it is called from.
> ls -l *.lic
-rw-r--r-- 1 user users 326 Apr 28 15:23 gurobi.lic
-rw-r--r-- 1 user users 1029 Mar 20 10:18 mosek.lic
-rw-r--r-- 1 user users 128 Apr 28 15:25 stata.lic
Set the tag and repo accordingly.
tag=2024-03-30
repo=aer-2023-0700
space=aeadataeditor
Default parameters are encoded in config.sh, which is called by all utility scripts.
source config.shIf using a pre-built image on Docker Hub,
docker pull ${space}/${repo}:${tag}
The image was preserved on Zenodo at 10.5281/zenodo.11080918, and can be obtained from there.
wget https://zenodo.org/record/11080918/files/${repo}-save.${tag}.tar
docker load -i ${repo}-save.${tag}.tar
Basic run of the image is possible with
docker run -it --rm ${space}/${repo}:${tag}
will run the base image (default entry point is to run Rstudio). However, it will not have access to the licensed software.
You will need all licenses available, and can then use run_docker.sh to obtain a shell.
bash ./run_docker.sh
or optionally
bash ./run_docker.sh (TAG)
to run with a tag that is different from the default encoded one.
bash ./run_docker.sh (TAG) (COMMAND)
is also possible.
The replication package for AER-2023-0700 has 8 scripts that each runs a part of the replication analysis.
| Script | Run in Docker? | Number of jobs | Software used | Notes |
|---|---|---|---|---|
execute-part1.sh |
Yes | 1 | Python, R, Mosek, Gurobi, Stata | We commented out modules and Python environment initialization, as all modules and Python packages are installed in the Docker image |
execute-part2.sh |
No | 1 | MATLAB, R | Base R is needed, but no additional packages. MATLAB uses parallel processing up to the number of physical cores, or the limits of the license, whichever is lower. |
execute-part3.sh |
Yes | 1 | R, Stata, Mosek | |
execute-part4.sh |
Yes | 100 | Python, Gurobi | We ran Docker within SLURM, limiting the number of parallel jobs to 2 (due to Gurobi licensing) |
execute-part5.sh |
Yes | 1 | Python, Stata |
Parts 6-8 were not run as part of the reproducibility checks, as they are MCMC runs that require additional adaptations to work in a particular computing environment. The container provided here is not optimized for speed (primarily due to Gurobi licensing restrictions), and the MCMC would have taken too long to run.
You can also run the Rstudio interface. You will need all licenses available, and can then use start_rstudio.sh to use Rstudio and/or terminal interactively. By default, Rstudio will run at https://localhost:8787, without a password.
bash ./start_rstudio.sh
or optionally
bash ./start_rstudio.sh (TAG)
to run with a Docker hub tag that is different from the default encoded one.
See Dockerfile for the image build instructions. At this time, the image uses
- rocker images as base images
- stata image for Stata provided the Social Science Data Editors
- Gurobi tar files provided by Gurobi
- Mosek tar files provided by Mosek
- Python installed from Ubuntu repositories
requirements.txtlists Python packages to be installedsetup.Rlists R packages to be installed- No Stata packages are installed at build time.
To build the image, use
bash ./build.sh (TAG)
You will need to manually push the image after it is built. The appropriate command line is printed out at the end of the build process.
To export the image, use
source config.sh
docker save -o ${repo}-save.${tag}.tar ${space}/${repo}:${tag}
All binary components remain copyright by their respective owners (Mosek, Gurobi, Stata, R, Python). The Dockerfile and scripts are released under the BSD license.