-
Notifications
You must be signed in to change notification settings - Fork 1
/
run_hparam.sh
executable file
·35 lines (32 loc) · 1.07 KB
/
run_hparam.sh
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
#!/usr/bin/env bash
set -e
run_job () {
JOBARRAY=($1)
MODEL=${JOBARRAY[0]}
DATASET=${JOBARRAY[1]}
PENALTY=${JOBARRAY[2]}
R=${JOBARRAY[3]}
LIFT_LAYERS=${JOBARRAY[4]}
SLOT=$2
echo $MODEL $DATASET $((SLOT - 1))
CUDA_VISIBLE_DEVICES=$((SLOT - 1)) python train.py \
--stepwise=True --steps=50000 \
--valid_freq=1000 --dropout=0 \
--prefix=230924_hparam_TU \
--model_type=$MODEL --dataset=$DATASET \
--num_layers=$LIFT_LAYERS --rank=$R --vc_penalty=$PENALTY --problem_type=vertex_cover \
--positional_encoding=laplacian_eigenvector --pe_dimension=$((R/2)) \
--batch_size=16
}
export -f run_job
for model in 'LiftMP' 'Nikos' ; do
for dataset in 'ENZYMES' 'PROTEINS' 'IMDB-BINARY' 'MUTAG' 'COLLAB' ; do
for penalty in '1' ; do
for r in '4' '8' '16' ; do
for lift_layers in '1' '4' '8' '12' ; do
echo $model $dataset $penalty $r $lift_layers
done
done
done
done
done | parallel --ungroup -j2 run_job {} {%}