| import os,time,json,collections |
| os.environ.setdefault('MPLCONFIGDIR',os.path.abspath('.cache/matplotlib')) |
| from pathlib import Path |
| import numpy as np,joblib |
| from inference import Predictor |
| from bci_data import cached |
| from evaluate import candidates |
|
|
| def run(): |
| groups={} |
| for s,r,p in candidates(): |
| if s!=1:continue |
| groups[(r['model'],r['montage'],r['channels'],r.get('seed','deterministic'))]=(r,p) |
| x,_,_=cached(1,'validation');out=[] |
| for key,(r,p) in sorted(groups.items(),key=lambda x:str(x[0])): |
| t=time.perf_counter();m=Predictor(p);loadtime=time.perf_counter()-t;m.predict(x[0]);times=[m.predict(x[i])['latency_ms'] for i in range(5)];extra={} |
| if m.neural:extra['parameters']=sum(q.numel() for q in m.model.parameters()) |
| elif hasattr(m.model,'steps'): |
| last=m.model.steps[-1][1] |
| if hasattr(last,'n_support_'):extra['support_vectors']=int(last.n_support_.sum());extra['feature_dimensions']=last.n_features_in_ |
| elif hasattr(last,'coef_'):extra['coefficients']=int(last.coef_.size) |
| out.append({'model':key[0],'montage':key[1],'channels':key[2],'seed':key[3],'representative_subject':1,'source_split':'validation','warm_trials':5,'median_preprocess_and_inference_ms':float(np.median(times)),'range_ms':[min(times),max(times)],'load_seconds':loadtime,'checkpoint_bytes':p.stat().st_size,'device':'cpu',**extra}) |
| Path('reports/inference_benchmarks.json').write_text(json.dumps(out,indent=2));print('Benchmarked',len(out),'configurations') |
| if __name__=='__main__':run() |
|
|
