Add high-fidelity training script (5K neurons, 2s sim, 200 epochs)

4h walltime per recording for proper convergence. Targets the 6
baseline Sharf organoid recordings. Previous fast runs achieved 79%
FR ratio; this should reach >95% with longer sims and more epochs.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: m9h

scripts/slurm_train_hifi.sh

@@ -0,0 +1,61 @@
+#!/usr/bin/env bash
+#SBATCH --job-name=bl1-hifi
+#SBATCH --partition=gpu
+#SBATCH --gres=gpu:1
+#SBATCH --cpus-per-task=8
+#SBATCH --mem=64G
+#SBATCH --time=04:00:00
+#SBATCH --output=/data/datasets/bl1/results/sharf_2022/hifi_logs/%x_%A_%a.out
+#SBATCH --error=/data/datasets/bl1/results/sharf_2022/hifi_logs/%x_%A_%a.err
+# ---------------------------------------------------------------------------
+# High-fidelity training: 5K neurons, 2000ms sim, 200 epochs
+#
+#   sbatch --array=0-9 scripts/slurm_train_hifi.sh    # 10 baseline recordings
+#   sbatch scripts/slurm_train_hifi.sh 0               # single recording
+# ---------------------------------------------------------------------------
+
+set -euo pipefail
+
+PROJECT_DIR="/home/mhough/dev/bl1"
+VENV="${PROJECT_DIR}/.venv/bin"
+DATA_DIR="/data/datasets/bl1/zenodo_sharf_2022"
+OUT_DIR="/data/datasets/bl1/results/sharf_2022/hifi"
+
+# Use baseline recordings (best-characterized, ~0.3-0.86 Hz)
+mapfile -t FILES < <(ls -1 "${DATA_DIR}"/7month_*.raw.h5 2>/dev/null | sort)
+
+IDX="${SLURM_ARRAY_TASK_ID:-${1:-0}}"
+if [[ $IDX -ge ${#FILES[@]} ]]; then
+    echo "Index $IDX out of range (${#FILES[@]} files)"
+    exit 1
+fi
+
+FILE="${FILES[$IDX]}"
+BASENAME=$(basename "$FILE" .raw.h5)
+
+echo "============================================================"
+echo "BL-1 High-Fidelity Training"
+echo "============================================================"
+echo "  Job ID:     ${SLURM_JOB_ID:-local}"
+echo "  Recording:  ${BASENAME}"
+echo "  Config:     5K neurons, 2000ms sim, 200 epochs"
+echo "  Date:       $(date)"
+echo "============================================================"
+
+mkdir -p "${OUT_DIR}/${BASENAME}"
+
+cd "${PROJECT_DIR}"
+
+"${VENV}/python" scripts/train_culture.py \
+    --n-neurons 5000 \
+    --n-epochs 200 \
+    --sim-duration-ms 2000 \
+    --lr 5e-5 \
+    --surrogate-beta 5.0 \
+    --auto-noise \
+    --from-recording "${FILE}" \
+    --output-dir "${OUT_DIR}/${BASENAME}"
+
+echo ""
+echo "Job complete: ${BASENAME}"
+echo "============================================================"