8f43927575
examples/ is based on KasperskyLab/knp at 55470ed6 with local changes to the mnist-learn altai/blifat models, cuda-backend CMake and model_loader. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
74 lines
3.8 KiB
Python
74 lines
3.8 KiB
Python
"""Run real KNP training/inference and preserve logs, settings and count-based metrics."""
|
|
import argparse
|
|
import csv
|
|
import io
|
|
import json
|
|
import os
|
|
from pathlib import Path
|
|
import subprocess
|
|
import time
|
|
|
|
ap = argparse.ArgumentParser()
|
|
ap.add_argument('name')
|
|
ap.add_argument('--dataset', default='validation')
|
|
ap.add_argument('--train', type=int, default=8000)
|
|
ap.add_argument('--count', type=int, default=2000)
|
|
ap.add_argument('--threshold', type=int, default=60)
|
|
ap.add_argument('--data-dir', type=Path, default=Path('/home/int/run/cats-dogs/data'))
|
|
ap.add_argument('--columns', type=int, default=20)
|
|
ap.add_argument('--seed', type=int, default=42)
|
|
ap.add_argument('--jitter', type=float, default=20)
|
|
ap.add_argument('--quant', choices=['default', 'weights'], default='default')
|
|
ap.add_argument('--backend', choices=['single', 'multi', 'tuned', 'fast'], default='single')
|
|
ap.add_argument('--threads', type=int, default=2)
|
|
a = ap.parse_args()
|
|
if a.backend == 'multi':
|
|
ap.error('Stock multi-threaded KNP has no AltAI types; use --backend tuned for the experimental AltAI port')
|
|
if not 1 <= a.threads <= 256:
|
|
ap.error('--threads must be between 1 and 256')
|
|
root = Path('/home/int/run/cats-dogs')
|
|
out = root / a.name
|
|
out.mkdir(exist_ok=False)
|
|
env = dict(os.environ, LD_LIBRARY_PATH='/home/int/knp-build/bin:/home/int/deps/lib',
|
|
KNP_COL_JITTER=str(a.jitter), KNP_COL_SEED=str(a.seed), KNP_COL_SIZE=str(a.columns),
|
|
KNP_MNIST_THR=str(a.threshold), KNP_MNIST_ADD_THR_SCALE='0',
|
|
KNP_MNIST_INFER_INHIB='255', KNP_MNIST_DUMP=str(out))
|
|
if a.quant == 'weights':
|
|
env['KNP_MNIST_QUANT'] = 'weights'
|
|
else:
|
|
env.pop('KNP_MNIST_QUANT', None)
|
|
env['KNP_CPU_THREADS'] = str(a.threads)
|
|
backends = {'single': '/home/int/knp-build/bin/libknp-cpu-single-threaded-backend.so',
|
|
'multi': '/home/int/knp-build/bin/libknp-cpu-multi-threaded-backend.so',
|
|
'tuned': str(root / 'acceleration/libknp-cpu-tuned-backend.so'),
|
|
'fast': str(root / 'fast-cpu/libknp-cpu-fast-backend.so')}
|
|
command = [str(root / 'build/cats_dogs'), '-m', 'altai',
|
|
'--images', str(a.data_dir / (a.dataset + '.bin')),
|
|
'--labels', str(a.data_dir / (a.dataset + '.target')),
|
|
'-t', str(a.train), '-i', str(a.count), '-l', 'error',
|
|
'--training_backend', backends[a.backend]]
|
|
settings = {'arguments': {k: str(v) if isinstance(v, Path) else v for k, v in vars(a).items()}, 'command': command,
|
|
'environment': {k: v for k, v in env.items() if k.startswith('KNP_')},
|
|
'started_utc': time.strftime('%Y-%m-%dT%H:%M:%SZ', time.gmtime())}
|
|
(out / 'settings.json').write_text(json.dumps(settings, indent=2) + '\n')
|
|
start = time.monotonic()
|
|
with (out / 'run.log').open('w') as log:
|
|
result = subprocess.run(command, input='\n', text=True, env=env, stdout=log, stderr=subprocess.STDOUT)
|
|
text = (out / 'run.log').read_text()
|
|
if result.returncode:
|
|
raise RuntimeError(f'KNP exited {result.returncode}: {text[-2000:]}')
|
|
lines = text[text.index('CLASS,TOTAL_VOTES'):].splitlines()[:3]
|
|
(out / 'result.csv').write_text('\n'.join(lines) + '\n')
|
|
rows = list(csv.DictReader(io.StringIO('\n'.join(lines))))
|
|
metrics = {'elapsed_seconds': round(time.monotonic() - start, 2), 'classes': {}}
|
|
for row in rows:
|
|
label = 'cat' if row['CLASS'] == '0' else 'dog'
|
|
metrics['classes'][label] = {'correct': int(row['TRUE_POSITIVES']),
|
|
'unanswered': int(row['FALSE_NEGATIVES']),
|
|
'wrong': int(row['FALSE_POSITIVES'])}
|
|
for key in ('correct', 'unanswered', 'wrong'):
|
|
metrics[key] = sum(row[key] for row in metrics['classes'].values())
|
|
assert metrics['correct'] + metrics['unanswered'] + metrics['wrong'] == a.count
|
|
metrics['accuracy'] = metrics['correct'] / a.count
|
|
(out / 'metrics.json').write_text(json.dumps(metrics, indent=2) + '\n')
|
|
print(json.dumps(metrics), flush=True)
|