ml4a-scheduling

Code accompanying our ICML 2023 paper: "On Preemption and Learning in Stochastic Scheduling" by Nadav Merlis^(*), Hugo Richard^(*), Flore Sentenac^(*), Corentin Odic, Mathieu Molina and Vianney Perchet

^(*): Equal contribution

Documentation: https://hugorichard.github.io/ml4a-scheduling/api.html

Install

Go in the ml4a-scheduling directory. After required dependency are installed (via pip install -r requirements.txt) run:

pip install -e .

Requirements

For the core algorithms:

• numpy>=1.16
• scipy>=1.3.2
• numba >= 0.56.4

To reproduce and plot the experiments (Figures 1, 2 and 3).

• matplotlib>=2.0.0
• joblib>=1.2.0

To run the tests:

• pytest>=5.0.1

To install all requirements, go in the ml4a-scheduling directory and run pip install -r requirements.txt

Experiments

Scheduling 2 groups of a varying number of jobs with fixed expected job processing time (Figure 1)

Go in the experiments directory.

Run python all_algos-vary-nsamples.py

Go in the plotting directory.

Run python plot_all_algos-vary-nsamples.py

See the figure in figures/all_algos_vary_nsamples.pdf.

Scheduling 2 groups of n=50 jobs with varying expected job processing time (Figure 2)

Go in the experiments directory.

Run python n_50_vary_lambda_logscale.py

Go in the plotting directory.

Run python plot_n_50_vary_lambda_logscale.py

See the figure in figures/n_50_vary_lambda_logscale.pdf.

Benchmark LSEPT with 2 groups of a varying number of jobs with fixed expected job processing time (Figure 3)

Go in the experiments directory.

Run python nsamples_lsept.py

Go in the plotting directory.

Run python plot_vary-nsamples_lsept.py

See the figure in figures/vary-nsamples_lsept.pdf.