"""Extraction-only target schema shared by conversion, training and evaluation.""" import json from rubric import NOTE_TYPES, NOT_APPLICABLE, items_for, missing_items, score_from_items, validate SCHEMA_VERSION='extraction-v1' def target_text(output): if not isinstance(output,dict) or set(output)!={'type','items'}: raise ValueError('Target must contain exactly type and items') _, _, errors = parse_extraction(output) if errors: raise ValueError('; '.join(errors)) return json.dumps(output,ensure_ascii=False,separators=(',',':')) def parse_extraction(text,note=None): errors=[] try: def unique(pairs): result={} for k,v in pairs: if k in result:raise ValueError('duplicate JSON key: '+k) result[k]=v return result obj=json.loads(text,object_pairs_hook=unique) if isinstance(text,str) else text if not isinstance(obj,dict) or set(obj)!={'type','items'}:raise ValueError('expected exactly type and items') t,it=obj['type'],obj['items'] if t not in NOTE_TYPES:raise ValueError('unknown type') if not isinstance(it,dict) or set(it)!=set(items_for(t)):raise ValueError('missing or extra item keys') if not all(isinstance(v,str) and bool(v.strip()) for v in it.values()):raise ValueError('items must be nonempty strings') na=NOT_APPLICABLE.get(t,set()) for k,v in it.items(): if k in na and v!='n/a':errors.append(k+' must be n/a') if k not in na and v=='n/a':errors.append(k+' cannot be n/a') if note is not None:errors+=validate(note,t,it) return t,it,errors except (ValueError,TypeError,KeyError) as e:return None,{},[str(e)] def derived_fields(output,note=None): t,it,errors=parse_extraction(output,note) if errors:raise ValueError('; '.join(errors)) return {'score':score_from_items(t,it),'missing_items':missing_items(t,it)}