|
| 1 | +from docarray import DocList, BaseDoc |
| 2 | +from docarray.typing import AnyTensor |
| 3 | +from pydantic import create_model |
| 4 | +from typing import Dict, List, Any, Union, Optional |
| 5 | + |
| 6 | + |
| 7 | +def _create_aux_model_doc_list_to_list(model): |
| 8 | + fields = {} |
| 9 | + for field_name, field in model.__annotations__.items(): |
| 10 | + try: |
| 11 | + if issubclass(field, DocList): |
| 12 | + fields[field_name] = (List[field.doc_type], {}) |
| 13 | + else: |
| 14 | + fields[field_name] = (field, {}) |
| 15 | + except TypeError: |
| 16 | + fields[field_name] = (field, {}) |
| 17 | + return create_model( |
| 18 | + model.__name__, __base__=model, __validators__=model.__validators__, **fields |
| 19 | + ) |
| 20 | + |
| 21 | + |
| 22 | +def _get_field_from_type( |
| 23 | + field_schema, |
| 24 | + field_name, |
| 25 | + root_schema, |
| 26 | + cached_models, |
| 27 | + is_tensor=False, |
| 28 | + num_recursions=0, |
| 29 | +): |
| 30 | + field_type = field_schema.get('type', None) |
| 31 | + tensor_shape = field_schema.get('tensor/array shape', None) |
| 32 | + if 'anyOf' in field_schema: |
| 33 | + any_of_types = [] |
| 34 | + for any_of_schema in field_schema['anyOf']: |
| 35 | + if '$ref' in any_of_schema: |
| 36 | + obj_ref = any_of_schema.get('$ref') |
| 37 | + ref_name = obj_ref.split('/')[-1] |
| 38 | + any_of_types.append( |
| 39 | + create_base_doc_from_schema( |
| 40 | + root_schema['definitions'][ref_name], |
| 41 | + ref_name, |
| 42 | + cached_models=cached_models, |
| 43 | + ) |
| 44 | + ) |
| 45 | + else: |
| 46 | + any_of_types.append( |
| 47 | + _get_field_from_type( |
| 48 | + any_of_schema, |
| 49 | + field_name, |
| 50 | + root_schema=root_schema, |
| 51 | + cached_models=cached_models, |
| 52 | + is_tensor=tensor_shape is not None, |
| 53 | + num_recursions=0, |
| 54 | + ) |
| 55 | + ) # No Union of Lists |
| 56 | + ret = Union[tuple(any_of_types)] |
| 57 | + for rec in range(num_recursions): |
| 58 | + ret = List[ret] |
| 59 | + elif field_type == 'string': |
| 60 | + ret = str |
| 61 | + for rec in range(num_recursions): |
| 62 | + ret = List[ret] |
| 63 | + elif field_type == 'integer': |
| 64 | + ret = int |
| 65 | + for rec in range(num_recursions): |
| 66 | + ret = List[ret] |
| 67 | + elif field_type == 'number': |
| 68 | + if num_recursions <= 1: |
| 69 | + # This is a hack because AnyTensor is more generic than a simple List and it comes as simple List |
| 70 | + if is_tensor: |
| 71 | + ret = AnyTensor |
| 72 | + else: |
| 73 | + ret = List[float] |
| 74 | + else: |
| 75 | + ret = float |
| 76 | + for rec in range(num_recursions): |
| 77 | + ret = List[ret] |
| 78 | + elif field_type == 'boolean': |
| 79 | + ret = bool |
| 80 | + for rec in range(num_recursions): |
| 81 | + ret = List[ret] |
| 82 | + elif field_type == 'object' or field_type is None: |
| 83 | + if 'additionalProperties' in field_schema: # handle Dictionaries |
| 84 | + additional_props = field_schema['additionalProperties'] |
| 85 | + if additional_props.get('type') == 'object': |
| 86 | + ret = Dict[ |
| 87 | + str, |
| 88 | + create_base_doc_from_schema( |
| 89 | + additional_props, field_name, cached_models=cached_models |
| 90 | + ), |
| 91 | + ] |
| 92 | + else: |
| 93 | + ret = Dict[str, Any] |
| 94 | + else: |
| 95 | + obj_ref = field_schema.get('$ref') or field_schema.get('allOf', [{}])[ |
| 96 | + 0 |
| 97 | + ].get('$ref', None) |
| 98 | + if num_recursions == 0: # single object reference |
| 99 | + if obj_ref: |
| 100 | + ref_name = obj_ref.split('/')[-1] |
| 101 | + ret = create_base_doc_from_schema( |
| 102 | + root_schema['definitions'][ref_name], |
| 103 | + ref_name, |
| 104 | + cached_models=cached_models, |
| 105 | + ) |
| 106 | + else: |
| 107 | + ret = Any |
| 108 | + else: # object reference in definitions |
| 109 | + if obj_ref: |
| 110 | + ref_name = obj_ref.split('/')[-1] |
| 111 | + ret = DocList[ |
| 112 | + create_base_doc_from_schema( |
| 113 | + root_schema['definitions'][ref_name], |
| 114 | + ref_name, |
| 115 | + cached_models=cached_models, |
| 116 | + ) |
| 117 | + ] |
| 118 | + else: |
| 119 | + ret = DocList[ |
| 120 | + create_base_doc_from_schema( |
| 121 | + field_schema, field_name, cached_models=cached_models |
| 122 | + ) |
| 123 | + ] |
| 124 | + elif field_type == 'array': |
| 125 | + ret = _get_field_from_type( |
| 126 | + field_schema=field_schema.get('items', {}), |
| 127 | + field_name=field_name, |
| 128 | + root_schema=root_schema, |
| 129 | + cached_models=cached_models, |
| 130 | + is_tensor=tensor_shape is not None, |
| 131 | + num_recursions=num_recursions + 1, |
| 132 | + ) |
| 133 | + else: |
| 134 | + if num_recursions > 0: |
| 135 | + raise ValueError( |
| 136 | + f"Unknown array item type: {field_type} for field_name {field_name}" |
| 137 | + ) |
| 138 | + else: |
| 139 | + raise ValueError( |
| 140 | + f"Unknown field type: {field_type} for field_name {field_name}" |
| 141 | + ) |
| 142 | + return ret |
| 143 | + |
| 144 | + |
| 145 | +def create_base_doc_from_schema( |
| 146 | + schema: Dict[str, any], model_name: str, cached_models: Optional[Dict] = None |
| 147 | +) -> type: |
| 148 | + cached_models = cached_models if cached_models is not None else {} |
| 149 | + fields = {} |
| 150 | + if model_name in cached_models: |
| 151 | + return cached_models[model_name] |
| 152 | + for field_name, field_schema in schema.get('properties', {}).items(): |
| 153 | + field_type = _get_field_from_type( |
| 154 | + field_schema=field_schema, |
| 155 | + field_name=field_name, |
| 156 | + root_schema=schema, |
| 157 | + cached_models=cached_models, |
| 158 | + is_tensor=False, |
| 159 | + num_recursions=0, |
| 160 | + ) |
| 161 | + fields[field_name] = (field_type, field_schema.get('description')) |
| 162 | + |
| 163 | + model = create_model(model_name, __base__=BaseDoc, **fields) |
| 164 | + cached_models[model_name] = model |
| 165 | + return model |
0 commit comments