#!/bin/bash
# ============================================================================
# run_paper_experiments.sh  (MINIMAL — reuse existing genuine models)
#
# All deep models (U-Net, Attention U-Net, SegFormer, Swin-UNet, Swin ensemble)
# are ALREADY trained and tested on the held-out test set — those results are
# genuine and are reused as-is (no retraining).
#
# The ONLY genuinely missing experiment is the classical Random-Forest baseline,
# which this script runs (CPU, fast). It then prints the complete results table.
# ============================================================================

source /data1/nasiruddink/miniconda3/etc/profile.d/conda.sh
conda activate brats
export TF_CPP_MIN_LOG_LEVEL=2
cd /data1/nasiruddink/rockglacier/RockGlacierNet

echo "############################################################"
echo "# Random-Forest baseline (classical ML on the 20 bands)     "
echo "############################################################"
if [ -f bandselection/rf_baseline/metrics.json ]; then
  echo "[skip] RF baseline already done"
else
  echo "[RF $(date +%H:%M)] training random forest (CPU, ~5-15 min)..."
  python3 rf_baseline.py 2>&1 | tail -20
fi

echo ""
echo "############################################################"
echo "# Collecting all results                                    "
echo "############################################################"
python3 collect_paper_results.py