File size: 2,894 Bytes
3c8d50e | 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 | """Reload the selected checkpoint and evaluate the held-out trajectories."""
import hashlib
import json
import sys
import time
from pathlib import Path
sys.path.insert(0,str(Path(__file__).resolve().parents[1]))
import numpy as np
import torch
from model.vrba import VRBADeepONet
from common import arguments,config,path,accelerator,load_data,metrics,write_json
def run(cfg,checkpoint=None):
torch.set_num_threads(1)
device=accelerator(cfg['training']['device'])
checkpoint=path(checkpoint or cfg['paths']['best_checkpoint'])
ck=torch.load(checkpoint,map_location='cpu',weights_only=False)
model=VRBADeepONet(**ck['model_config']).to(device)
model.load_state_dict(ck['model_state_dict'],strict=True); model.eval()
if cfg['data']['sensors']!=ck['model_config']['sensors']:raise ValueError('Pressure sensor count mismatch.')
data=load_data(cfg,stats=ck['normalization'])
s=data['splits']['test']; x=torch.as_tensor(data['x'][s],device=device)
t=torch.as_tensor(data['time'],device=device)
begin=time.perf_counter(); torch.cuda.synchronize()
with torch.inference_mode():
pred=torch.cat([model(x[i:i+50],t) for i in range(0,len(x),50)]).cpu().numpy()
torch.cuda.synchronize(); duration=time.perf_counter()-begin
truth=data['radius'][s]; pred_physical=pred.astype(np.float64)*ck['normalization']['radius_std']+ck['normalization']['radius_mean']
per_case=np.linalg.norm(pred_physical-truth,axis=1)/np.linalg.norm(truth,axis=1)
normalized=metrics(pred,data['y'][s]); physical=metrics(pred_physical,truth)
result=path(cfg['paths']['results']);result.mkdir(parents=True,exist_ok=True)
np.savez_compressed(result/'predictions.npz',prediction=pred_physical,target=truth,
prediction_normalized=pred,target_normalized=data['y'][s],time=data['output_t'],
pressure=data['pressure'][s],pressure_time=data['pressure_t'],
original_indices=np.arange(s.start,s.stop),per_trajectory_relative_l2=per_case)
report=dict(checkpoint_name=checkpoint.name,checkpoint_sha256=hashlib.sha256(checkpoint.read_bytes()).hexdigest(),
checkpoint_epoch=ck['epoch'],best_validation=ck['best_validation'],test_count=len(x),
inference_seconds=duration,physical=physical,normalized=normalized,
paper_reference=dict(relative_l2=0.00516,residual_variance=1.94e-5,residual_linf=0.0822),
comparability='Different reproducible split; authors original indices unknown. Report both physical and train-normalized metrics; table aggregation is not fully specified.',
worst_original_index=int(s.start+np.argmax(per_case)))
write_json(result/'metrics.json',report)
print('TEST_METRICS '+json.dumps(report),flush=True)
return report
if __name__=='__main__':
p=arguments(__doc__);p.add_argument('--checkpoint',default=None)
a=p.parse_args();run(config(a.config),a.checkpoint)
|