add: final aggregation utils to obtain one dataset only

examples/behavior_1k/aggregate.slurm

@@ -0,0 +1,20 @@
+#!/bin/bash
+#SBATCH -J b1k-aggregate
+#SBATCH -p hopper-cpu
+#SBATCH --qos=high
+#SBATCH -c 2              # More CPUs won't help much (I/O bound)
+#SBATCH -t 20:00:00       # Conservative: 5.2x pattern suggests ~13h, so 12h is tight but reasonable
+#SBATCH --mem=8G          # Peak will be ~5-6GB, 8G gives comfortable margin
+#SBATCH -o logs/aggregate-50-%j.out
+#SBATCH -e logs/aggregate-50-%j.err
+
+set -euo pipefail
+export OMP_NUM_THREADS=${SLURM_CPUS_PER_TASK:-1}
+export PYTHONUNBUFFERED=1
+
+source "$HOME/.bashrc"
+conda activate lerobot
+
+cd /admin/home/francesco_capuano/lerobot
+
+python examples/behavior_1k/aggregate_tasks_datasets.py

examples/behavior_1k/aggregate_tasks_datasets.py

@@ -0,0 +1,30 @@
+from pathlib import Path
+
+from lerobot.datasets.aggregate import aggregate_datasets
+from lerobot.datasets.lerobot_dataset import LeRobotDataset
+
+
+def main():
+    """Aggregate all tasks datasets into a single LeRobotDataset and push it to the hub."""
+    task_indices = range(50)
+
+    repo_ids = [f"fracapuano/behavior1k-task{i:04d}" for i in task_indices]
+
+    roots = [Path(f"/fsx/francesco_capuano/behavior1k/behavior1k-task{i:04d}") for i in task_indices]
+
+    aggregated_root = Path("/fsx/francesco_capuano/behavior1k/behavior1k")
+    aggregated_repo_id = "fracapuano/behavior1k"
+
+    aggregate_datasets(
+        repo_ids=repo_ids,
+        roots=roots,
+        aggr_repo_id=aggregated_repo_id,
+        aggr_root=aggregated_root,
+    )
+
+    ds = LeRobotDataset(repo_id=aggregated_repo_id, root=aggregated_root)
+    ds.push_to_hub()
+
+
+if __name__ == "__main__":
+    main()