experiments

Instructions on running the experiments

Environments

The experiments require Conda to be installed in the system. Two virtual environments can be prepared by running the following two commands.

conda create --name=cvar_sensing python=3.10 --yes
conda create --name=asac_sensing python=3.7.10 --yes

Install the package

For most of the experiments, we use the first environment.

conda activate cvar_sensing
pip install "cvar_sensing[benchmarks] @ git+https://github.com/yvchao/cvar_sensing.git"
# OR if you have the repository cloned locally, from inside the repository directory run:
pip install .[benchmarks]

The baseline of ASAC requires TensorFlow 1.5. We install its dependencies in a separate environment.

conda activate asac_sensing
pip install "cvar_sensing[asac] @ git+https://github.com/yvchao/cvar_sensing.git"
# OR if you have the repository cloned locally, from inside the repository directory run:
pip install .[benchmarks]

Check experiment configs

Open exp_config.py and update the values for

• venv
• venv_asac
• _conda_path

if necessary.

Run the experiment

All experiments will be conducted by running the script of "./run_experiment.py".

conda activate base
python ./run_experiment.py

Note

ASAC relies on an older version of TensorFlow. The results from the ASAC baseline are unstable even with the same random seeds. Currently we find no treat for this issue.

Check the results

Analysis of the experimental results can be generated and examined with files in the reports folder.