#!/bin/bash
#SBATCH --job-name=sample-spice-subsets
#SBATCH --output=slurm-%x-%j.out
#SBATCH --time=24:00:00
#SBATCH --nodes=1
#SBATCH --ntasks-per-node=1
#SBATCH --gpus=1

set -euo pipefail

nvidia-smi

SCRIPT_DIR="/home/s5f/ev333.s5f/work/spice"
PYTHON_SCRIPT="$SCRIPT_DIR/sample_nested_subsets.py"
DATA_DIR="$SCRIPT_DIR/data"

source ~/.bashrc
conda activate mace

cd "$SCRIPT_DIR"

echo "=========================================="
echo "SPICE Dataset Nested Subsampling"
echo "Using Furthest Point Sampling (FPS)"
echo "=========================================="
echo ""

# Sample training set
echo "Processing training set: train_large_neut_no_bad_clean.xyz"
echo "Expected: ~951,005 structures"
echo ""

srun python3 "$PYTHON_SCRIPT" \
    --input "$DATA_DIR/train_large_neut_no_bad_clean.xyz" \
    --percentages 50 20 10 5 1 \
    --device cuda \
    --seed 42

echo ""
echo "=========================================="
echo ""

# Sample test set
echo "Processing test set: test_large_neut_all.xyz"
echo "Expected: ~50,195 structures"
echo ""

srun python3 "$PYTHON_SCRIPT" \
    --input "$DATA_DIR/test_large_neut_all.xyz" \
    --percentages 50 20 10 5 1 \
    --device cuda \
    --seed 42

echo ""
echo "=========================================="
echo "All subsets created successfully!"
echo "=========================================="