| from pathlib import Path |
| import json,sys |
| import numpy as np |
| import stim |
| sys.path.insert(0,'/workspace/research/exact_d3') |
| from exact import make_plan,parse_errors |
|
|
|
|
| def load_all(max_ops=3072): |
| root=Path('/workspace/data') |
| rows=[r for r in json.loads((root/'train/manifest.json').read_text())['conditions'] if r['distance']==3] |
| groups=sorted({r['group'] for r in rows});result=[] |
| for row in rows: |
| dem=stim.DetectorErrorModel.from_file(root/'train'/row['noise']);plan=make_plan(dem) |
| assert int(plan['width'])<=10 |
| count=len(plan['masks']);assert count<=max_ops,(row['name'],count) |
| coords=dem.get_detector_coordinates() |
| circuit=stim.Circuit.from_file(root/'train'/row['circuit']) |
| qubit_coords=circuit.get_final_qubit_coordinates() |
| first_measurement=next(ins for ins in circuit.flattened() if ins.name in ['M','MX','MR','MRX']) |
| spaces=sorted(tuple(qubit_coords[t.value][:2]) for t in first_measurement.targets_copy()) |
| assert len(spaces)==8 |
| idx={(int(c[2]),tuple(c[:2])):d for d,c in coords.items()} |
| errors=parse_errors(dem) |
| context=np.full((max_ops,40),200,np.int32) |
| static=np.zeros((max_ops,57),np.float32) |
| for op,source in enumerate(plan['source_indices']): |
| if source<0:continue |
| ds,logical,p=errors[source] |
| t=min(int(coords[d][2]) for d in ds) if ds else 0 |
| for dt in range(-2,3): |
| for s,spatial in enumerate(spaces):context[op,(dt+2)*8+s]=idx.get((t+dt,spatial),200) |
| for d in ds: |
| c=coords[d];static[op,(int(c[2])-t+2)*8+spaces.index(tuple(c[:2]))]=1 |
| static[op,40]=np.log(max(p,1e-7)/(1-max(p,1e-7)))/8 |
| static[op,41]=logical;static[op,42]=len(ds)/4 |
| static[op,43]=t/25;static[op,44]=row['rounds']/25 |
| static[op,45]=t==0;static[op,46]=t>=row['rounds']-1 |
| static[op,47+groups.index(row['group'])]=1 |
| static[op,55]=row['basis']=='X';static[op,56]=1 |
| masks=np.pad(plan['masks'],(0,max_ops-count)) |
| measured=np.pad(plan['measured'],(0,max_ops-count),constant_values=-1) |
| p=np.pad(plan['probabilities'],(0,max_ops-count)) |
| valid=((p>0)&(measured<0)).astype(np.float32) |
| logits=np.log(np.clip(p,1e-7,1-1e-7)/np.clip(1-p,1e-7,1-1e-7)).astype(np.float32) |
| item={'row':row,'context':context,'static':static,'masks':masks,'measured':measured,'base_logits':logits,'valid':valid} |
| for split in ['train','validation']: |
| a=np.load(root/split/row['arrays']);ds=np.unpackbits(a['detectors'],axis=1,count=row['num_detectors'],bitorder='little') |
| item[split+'_dets']=np.pad(ds,((0,0),(0,201-ds.shape[1]))) |
| item[split+'_labels']=a['labels'] |
| result.append(item) |
| return result |
|
|
