Arguments

$ARGUMENTS should contain: <dataset> [target]

•dataset (required): hcrl_sa, hcrl_ch, set_01, set_02, set_03, set_04
•target (optional): students_nokd, teachers, students, or a specific run like teacher_fusion

Parse the dataset and target from $ARGUMENTS. If only one word is provided, it is the dataset and no target filter is applied.

Usage Examples

code

/run-pipeline hcrl_sa students_nokd     # Run student without KD for hcrl_sa
/run-pipeline hcrl_ch teachers          # Run teacher pipeline for hcrl_ch
/run-pipeline set_01                    # Run all targets for set_01

Execution Steps

•
Parse arguments from $ARGUMENTS into dataset and optional target.
•
Verify dataset exists
bash
```
ls data/automotive/<dataset>/
```
•
Dry run first to see what will be executed. If a target is given, use it as a Snakemake rule name or build the output path experimentruns/<dataset>/<target>/best_model.pt. If no target, run the full DAG for that dataset.
bash
```
snakemake -s pipeline/Snakefile --config "datasets=[\"<dataset>\"]" -n 2>&1 | head -50
```

•

Submit to SLURM if dry run looks correct

bash

mkdir -p slurm_logs
snakemake -s pipeline/Snakefile --config "datasets=[\"<dataset>\"]" --profile profiles/slurm

•
Report the submitted job IDs and show how to monitor with squeue -u $USER.

Common Targets

Target	Description	Output
`teacher_fusion`	Full teacher pipeline	3 stages
`student_fusion`	Full student no-KD pipeline	3 stages
`student_fusion_kd`	Full student with KD pipeline	3 stages (needs teacher)
`teachers`	All datasets, teacher	Snakemake rule target
`students_nokd`	All datasets, student no-KD	Snakemake rule target
`students`	All datasets, student with KD	Snakemake rule target

Notes

•Pipeline runs on SLURM with GPU resources (V100, 128GB RAM)
•SLURM logs: slurm_logs/{jobid}-{rule}.{out,err}
•MLflow tracking is automatic
•Always do a dry run (-n) before submitting