#!/usr/bin/env bash
# Launch the official CEC-2022 sweep as many single-threaded shards across
# cores. Runs are independent and the per-combo resume-skip in
# run_cec2022_full.py makes disjoint (algo, seed) shards safe to run
# concurrently into one dir.
set -u
cd /home/admin1/Documents/sonthh/AHD_CMA
source .venv/bin/activate
OUT=outputs/runs/cec2022_rotated
LOG=outputs/logs
mkdir -p "$LOG" "$OUT"

# One BLAS thread per shard: CEC functions are tiny (10-20 dim), so
# multi-threaded BLAS gives no speedup but oversubscribes the cores.
export OMP_NUM_THREADS=1
export OPENBLAS_NUM_THREADS=1
export MKL_NUM_THREADS=1
export NUMEXPR_NUM_THREADS=1
export VECLIB_MAXIMUM_THREADS=1

ALGOS=(ahdcma bipop bohb cmaes doa gego grid gwo hyperband ihaho ipop optuna pso random scso woa)
# Two seed halves per algo -> 32 shards on 56 cores (leaves headroom for
# the BLAS-thread + OS). Each shard runs single-threaded.
SEEDS_A="0 1 2 3 4 5 6 7 8 9 10 11 12 13 14"
SEEDS_B="15 16 17 18 19 20 21 22 23 24 25 26 27 28 29"

n=0
for algo in "${ALGOS[@]}"; do
  for half in A B; do
    seeds=$([ "$half" = A ] && echo "$SEEDS_A" || echo "$SEEDS_B")
    nohup python scripts/run_cec2022_full.py --algos "$algo" --seeds $seeds \
      --output "$OUT" > "$LOG/cec_${algo}_${half}.log" 2>&1 &
    n=$((n+1))
  done
done
echo "launched $n single-threaded shards (16 algos x 2 seed-halves)"
wait
echo "ALL_CEC_SHARDS_DONE"