"""One single-thread HPO trial or fresh final refit; launch only through the CPU guard.""" import argparse import json import os from pathlib import Path import time import traceback import joblib import sklearn from sklearn.ensemble import HistGradientBoostingRegressor from threadpoolctl import threadpool_limits,threadpool_info from freeze_g1_components import sha from s2_recovery_supervisor import save from s2_models.features import histgbrt_grid,R2_COLUMNS from s2_r2_data import load_cell,learning_arrays,validation_score def run(root,index): from run_s2_r2 import contract start=time.monotonic();cpu_start=time.process_time();c=contract(root);job=c['jobs'][index] dest=root/'jobs'/f'job-{index:03d}'/'output';dest.mkdir(exist_ok=False) save(dest/'started.json',dict(unix=time.time(),pid=os.getpid(),job=job,contract_sha256=sha(root/'contract.json'),cpu_affinity=sorted(os.sched_getaffinity(0)))) try: assert sklearn.__version__=='1.7.2' config_index=job.get('config_index');selection=None if job['kind']=='FINAL': path=root/'selections'/f"cell-{job['cell']:02d}.json";selection=json.loads(path.read_text()) assert selection['status']=='SELECTED' and selection['cell']==job['cell'] config_index=selection['config_index'] for item in selection['trials']: assert sha(item['receipt_path'])==item['receipt_sha256'] save(dest/'selection.json',dict(path=str(path),sha256=sha(path),**selection)) params=histgbrt_grid()[config_index] with threadpool_limits(limits=1): pools=threadpool_info();assert all(p['num_threads']==1 for p in pools) learning,validation,roles=load_cell(root,c,job) save(dest/'access.json',dict(fold=job['fold'],pool=job['pool'],kind=job['kind'], cases=[dict(case=p['case'],role=p['role'],label_mode=p['label_mode'],label_sha256=p['label_sha256']) for p in learning+validation], outer_fold_test_labels_read=False,learning_only_final_refit=job['kind']=='FINAL')) x,y,w=learning_arrays(learning);load_cpu=time.process_time()-cpu_start save(dest/'progress.json',dict(stage='fit',training_queries=len(y),training_parts=len(learning),cpu_seconds=time.process_time()-cpu_start)) model=HistGradientBoostingRegressor(**params) fit_start=time.process_time();model.fit(x,y,sample_weight=w);fit_cpu=time.process_time()-fit_start assert model.n_iter_==params['max_iter'] value=errors=None if job['kind']=='HPO': save(dest/'progress.json',dict(stage='validation',fit_cpu_seconds=fit_cpu)) value,errors=validation_score(model,validation);save(dest/'validation-parts.json',errors) joblib.dump(model,dest/'model.joblib',compress=3) result=dict(status='COMPLETE',job=job,config_index=config_index,params=params,sklearn_version=sklearn.__version__, columns=list(R2_COLUMNS),target='T/t_set',learning_queries=len(y),learning_parts=len(learning), learning_components=len({p['component_id'] for p in learning}),validation_parts=len(validation), sample_weight_sum=float(w.sum()),validation_component_equal_relative_l2=value,fit_cpu_seconds=fit_cpu, load_cpu_seconds=load_cpu,worker_elapsed_cpu_seconds=time.process_time()-cpu_start,wall_seconds=time.monotonic()-start, threadpools=pools,cpu_affinity=sorted(os.sched_getaffinity(0)),model_sha256=sha(dest/'model.joblib'), roster_sha256=c['roster_sha256'],test_predictions=False,test_labels_read=False,random_state=0,retry=False, final_refit_learning_only=job['kind']=='FINAL',hpo_finished_configurations=selection['finished_configurations'] if selection else None) save(dest/'result.json',result);save(dest/'progress.json',dict(stage='complete',cpu_seconds=time.process_time()-cpu_start)) print(json.dumps(result),flush=True);return 0 except Exception as exc: save(dest/'failure.json',dict(status='FAILED_NO_RETRY',type=type(exc).__name__,message=str(exc),traceback=traceback.format_exc(), cpu_seconds=time.process_time()-cpu_start,wall_seconds=time.monotonic()-start)) raise if __name__=='__main__': p=argparse.ArgumentParser();p.add_argument('root',type=Path);p.add_argument('index',type=int);a=p.parse_args();raise SystemExit(run(a.root,a.index))