This repository contains the scripts that generated the figures used within the paper in addition to scripts used to replicate
experiments from the paper. There are two directories. First, the paper_figures/ directory includes original logs used to generate
the figures along with the scripts to generate them. Within each directory named figure_X_FIGURE_NAME/ exists logs/ and graphs/ directories,
containing the logs and the graphs generated respectively. Within each directory, exists at least one script named plot*.py. Running python plot*.py
generates the respective graph and writes the graph to the graphs/ directory. In addition, there are extra graphs not used in the paper that are included in some graphs/ directory.
To execute plot.py correctly, first run the command ./uncomporess_large_logs.sh which unzips large logs contained in some subdirectories.
The second directory is named experiments/. This directory contains scaled down experiments of the applications ran within the paper.
In this directory, there are two main experiments. DV3/ and RS-TriPhoton/. Detail on running these experiments is provided below:
The environments used to execute these applications are conda environments where instructions for installation can be found here: https://conda.io/projects/conda/en/latest/user-guide/install/index.html
Once Conda is installed the environment for the respective experiment can be created via the env.yml file located in each experiment's directory.
Within each experiment directory, execute the following command:
conda env create --name <ENVIRONMENT_NAME> --file=env.yml
This installs the environment needed to execute the environment, version numbers are shown within the YAML file:
Once installed, activate the environment with the following command:
conda activate <ENVIRONMENT_NAME>
To ensure distribution of the environment across workers within a cluster, we package the environment within a tarball using the following command:
poncho_package_create $CONDA_PREFIX <dv3-env|rstri-env>.tar.gz
This tarball is distributed along with worker binaries to ensure environments are available on remote execution sites.
Within each experiment directory exists a run.sh script which will execute the experiments needed to generate the example logs. Executed like so:
./run.sh
After completion, logs for each run can be found within the directory experiments/logs
After completion of both experiments, the logs can be processed to generate graphs similar to those shown in the paper. This can be done like so for each plotting script:
python plotting_tools/plot_fig_X.py
OR
./gen_graphs.sh
For all graphs at the same time
A graph will then be generated within the graphs/ directory.
More details on expected results and expected runtimes can be found within the pdf AD_SC24.pdf