TensorOpt4Entanglement — quick setup & run

Short instructions to install Julia packages used by this project and to run jobs using jobs.sh and a Slurm array (run.slurm).

Prerequisites

• Linux machine with Julia 1.11.x installed (the project was tested with Julia 1.11.6) and Mosek.
• A working Slurm cluster.
• Project directory layout:
  • benchmark/ — input instances
  • results/ — output directory
  • jobs.sh — job list generator (already provided)

Install required Julia packages

From the project root (this repo), prefer using the project environment if present:

julia --project=. -e 'using Pkg; Pkg.instantiate()'

This will install packages listed in Project.toml/Manifest.toml if they exist.

Prepare and run jobs locally

1. Edit the top of jobs.sh to set up experiments:

# at the top of jobs.sh (example)
algorithms=("LD1" "LDR0" "LDR1" "LDR2" "LDR3" "LDR4" "LDR5" "LDL" "D" "A")
timelimit=-1
datapath="$PWD/benchmark"
resultpath="$PWD/results"
juliabin="julia"

Example: automatic timelimit (-1) and auto-detect Julia executable.

• Use timelimit=-1 to let jobs.sh pick a sensible timeout per instance (based on filename/size).
• Detect the Julia binary at runtime so juliabin points to the actual executable found on PATH.

2. Run jobs.sh with bash from the project root to create job lists:

# simple run (uses defaults inside jobs.sh)
bash jobs.sh
# job_list.txt will be created in the project root

3. Set up Slurm environment (see run.slurm):

# Example Slurm header for distributed job array
#SBATCH --job-name=distributed_jobs
#SBATCH --ntasks=1
#SBATCH --cpus-per-task=1
#SBATCH --mem=10G
#SBATCH --time=05:58:00
#SBATCH --partition=opt_int
#SBATCH --constraint=Gold5222   # hardware feature or tag (e.g. "Gold5222")
#SBATCH --output=/dev/null      # redirect stdout (set to a file for debugging)
#SBATCH --error=/dev/null       # redirect stderr (set to a file for debugging)

Notes

• --job-name: human-readable job name.
• --ntasks: total MPI/tasks (1 for single-task jobs; increase for multi-task runs).
• --cpus-per-task: threads per task (set to number of threads your Julia worker uses).
• --mem: memory per node (or per job depending on cluster config). Adjust to workload.
• --time: wall-clock limit (format HH:MM:SS). Jobs exceeding this are killed.
• --partition: target partition/queue on the cluster.
• --constraint: node feature/label filter (matches nodes with this property).
• --output / --error: currently discarding logs to /dev/null; for debugging replace with a path like results/%x-%j.out and results/%x-%j.err.

4. Set up system environment variables used by runjobs.sh:

export JULIA_DEPOT_PATH=".julia_depot"
export MOSEKHOME=
export MOSEKLM_LICENSE_FILE=

5. Run all the experiments:

bash runjobs.sh

Run the Python script to parse result data

Prerequisites:

• Python 3.8+ installed.
• (Optional) Create and activate a virtual environment and install dependencies if a requirements file exists.

Basic usage (from project root):

# simple run (reads/writes paths relative to project root)
python3 ./exerpiementanaylysis.py

This prints the tables aggregating the results.