run_rruff.sh

#!/bin/bash
trap "exit" INT
raman_type=${1:-excellent_unoriented}
repeat_g=${2?:Error: the repeat version}
dir2save_ckpt=${3:-exp_data}  # USER-DEFINE
dir2load_data=${4:-data_group}  # USER-DEFINE
# ver=${5?:Error: what is the version}

lr_init=0.00005
lr_schedule=cosine
stem_kernel=21
stem_max_dim=64
depth=128
dist_aggregate=wave_channel_dot_L1

dropout=true
within_dropout=true
l2_regu_para=0.001

balanced=true
separable_act=true
balanced_alpha=1.0 #0.5
augmentation_mode=history

within_dropout=True
dropout=True
random_leave_one_out=True

ver=0
siamese_version=0

if [ "$raman_type" = excellent_unoriented ]; then
  max_epoch=2
  batch_size=81
elif [ "$raman_type" = raw ]; then
  max_epoch=3
  batch_size=83
fi

for repeat in $repeat_g
do
  python3 train_inception.py --version "$ver" --dataset RRUFF --raman_type "$raman_type" \
                --batch_size "$batch_size" --max_epoch "$max_epoch" \
                --learning_rate_init "$lr_init" \
                --distance_aggregation "$dist_aggregate" \
                --stem_kernel "$stem_kernel" \
                --lr_schedule "$lr_schedule" --l2_regu_para "$l2_regu_para" \
                --depth "$depth" --dropout "$dropout" \
                --neg_shift_scale 5 --stem_max_dim "$stem_max_dim" --within_dropout "$within_dropout" \
                --balanced "$balanced" --alpha "$balanced_alpha" \
                --siamese_version "$siamese_version" \
                --random_leave_one_out "$random_leave_one_out" --separable_act "$separable_act" \
                --augmentation_mode "$augmentation_mode" \
                --check_augmentation_on_datasplit false \
                --check_distance_on_datasplit false --repeat_on_python false \
                --repeat_time "$repeat" \
                --dir2load_data "$dir2load_data" --dir2save_ckpt "$dir2save_ckpt"
done