File size: 3,007 Bytes
f9ed2b4 | 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 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 | 'Plan or execute binary, source-classification and real-ratio training configurations.'
import argparse
import json
from pathlib import Path
import subprocess
import sys
ROOT = Path(__file__).resolve().parents[2]
SEEDS = [7, 13, 21, 42, 84]
N_GRID = [900, 90, 45, 18]
def build_jobs(kind, fixed_split, output, mapping, real_ratio=0.1):
if kind == 'ratio':
return [job for ratio in [0.01, 1.0] for job in build_jobs('ratio_binary', fixed_split, output / f'ratio_{ratio}', mapping, ratio)]
jobs = []
for seed in SEEDS if kind == 'binary' else [42]:
splits = output / 'splits' / f'seed{seed}'
jobs.append([sys.executable, str(ROOT/'scripts/data/build_training_splits.py'),
'--ai-train', str(fixed_split/'ai_train_pool.csv'), '--ai-val', str(fixed_split/'ai_validation.csv'),
'--real-train', str(fixed_split/'real_train_pool.csv'), '--real-val', str(fixed_split/'real_validation.csv'),
'--output-dir', str(splits), '--n-grid', ','.join(map(str,N_GRID)),
'--real-ratio', str(real_ratio), '--seed', str(seed)])
tag = 'training_split' if kind != 'source' else 'source_split'
split = splits/f'{tag}_n900_realratio_{str(real_ratio).replace(".", "p")}_seed{seed}.json'
for model in ['univfd','cnndetect']:
script=ROOT/f'train_{model}{"_multiclass" if kind=="source" else ""}.py'
for n in ([18, 900] if kind == 'ratio_binary' else N_GRID):
run=output/'runs'/f'{kind}_{model}'
if seed!=42: run=run/f'seed{seed}'
run=run/f'n{n}'
jobs.append([sys.executable,str(script),'--split-file',str(split),'--mapping',str(mapping),
'--sample-sizes',str(n),'--total-samples-per-model','9000','--train-ratio','1.0',
'--real-ratio',str(real_ratio),'--batch-size','64','--lr','0.0001',
'--num-validations','10','--seed',str(seed),'--output-dir',str(run)])
return jobs
def main():
p=argparse.ArgumentParser(description=__doc__)
p.add_argument('--kind',choices=['binary','source','ratio'],default='binary')
p.add_argument('--split-dir',type=Path,required=True)
p.add_argument('--mapping',type=Path,required=True)
p.add_argument('--output-dir',type=Path,default=ROOT/'outputs')
p.add_argument('--execute',action='store_true')
a=p.parse_args()
jobs=build_jobs(a.kind,a.split_dir.resolve(),a.output_dir.resolve(),a.mapping.resolve())
a.output_dir.mkdir(parents=True,exist_ok=True)
(a.output_dir/f'{a.kind}_release_jobs.json').write_text(json.dumps({
'protocol':'release-v1-resampled-final',
'checkpoint':'final','commands':jobs},indent=2),encoding='utf-8')
if a.execute:
for job in jobs: subprocess.run(job,cwd=ROOT,check=True)
print(f'{len(jobs)} commands {"executed" if a.execute else "planned (not executed)"}')
if __name__=='__main__':main()
|