[PLT-1463] Remove deserialize completely (#1818)

Author: Gabefire

.github/workflows/lbox-develop.yml

@@ -2,9 +2,9 @@ name: LBox Develop
 
 on:
   push:
-    branches: [develop]
+    branches: [develop, v6]
   pull_request:
-    branches: [develop]
+    branches: [develop, v6]
 
 concurrency:
   group: ${{ github.workflow }}-${{ github.ref }}

.github/workflows/python-package-develop.yml

@@ -2,9 +2,9 @@ name: Labelbox Python SDK Staging (Develop)
 
 on:
   push:
-    branches: [develop]
+    branches: [develop, v6]
   pull_request:
-    branches: [develop]
+    branches: [develop, v6]
 
 concurrency:
   group: ${{ github.workflow }}-${{ github.ref }}

libs/labelbox/src/labelbox/data/serialization/ndjson/base.py

@@ -8,18 +8,6 @@
 
 from ....annotated_types import Cuid
 
-subclass_registry = {}
-
-
-class _SubclassRegistryBase(BaseModel):
-    model_config = ConfigDict(extra="allow")
-
-    def __init_subclass__(cls, **kwargs):
-        super().__init_subclass__(**kwargs)
-        if cls.__name__ != "NDAnnotation":
-            with threading.Lock():
-                subclass_registry[cls.__name__] = cls
-
 
 class DataRow(_CamelCaseMixin):
     id: Optional[str] = None

libs/labelbox/src/labelbox/data/serialization/ndjson/classification.py

@@ -30,7 +30,6 @@
     model_serializer,
 )
 from pydantic.alias_generators import to_camel
-from .base import _SubclassRegistryBase
 
 
 class NDAnswer(ConfidenceMixin, CustomMetricsMixin):
@@ -224,7 +223,7 @@ def from_common(
 # ====== End of subclasses
 
 
-class NDText(NDAnnotation, NDTextSubclass, _SubclassRegistryBase):
+class NDText(NDAnnotation, NDTextSubclass):
     @classmethod
     def from_common(
         cls,
@@ -249,9 +248,7 @@ def from_common(
         )
 
 
-class NDChecklist(
-    NDAnnotation, NDChecklistSubclass, VideoSupported, _SubclassRegistryBase
-):
+class NDChecklist(NDAnnotation, NDChecklistSubclass, VideoSupported):
     @model_serializer(mode="wrap")
     def serialize_model(self, handler):
         res = handler(self)
@@ -298,9 +295,7 @@ def from_common(
         )
 
 
-class NDRadio(
-    NDAnnotation, NDRadioSubclass, VideoSupported, _SubclassRegistryBase
-):
+class NDRadio(NDAnnotation, NDRadioSubclass, VideoSupported):
     @classmethod
     def from_common(
         cls,
@@ -343,7 +338,7 @@ def serialize_model(self, handler):
         return res
 
 
-class NDPromptText(NDAnnotation, NDPromptTextSubclass, _SubclassRegistryBase):
+class NDPromptText(NDAnnotation, NDPromptTextSubclass):
     @classmethod
     def from_common(
         cls,

libs/labelbox/src/labelbox/data/serialization/ndjson/converter.py

@@ -26,20 +26,6 @@
 
 
 class NDJsonConverter:
-    @staticmethod
-    def deserialize(json_data: Iterable[Dict[str, Any]]) -> LabelGenerator:
-        """
-        Converts ndjson data (prediction import format) into the common labelbox format.
-
-        Args:
-            json_data: An iterable representing the ndjson data
-        Returns:
-            LabelGenerator containing the ndjson data.
-        """
-        data = NDLabel(**{"annotations": copy.copy(json_data)})
-        res = data.to_common()
-        return res
-
     @staticmethod
     def serialize(
         labels: LabelCollection,

libs/labelbox/src/labelbox/data/serialization/ndjson/label.py

@@ -46,7 +46,6 @@
 from .relationship import NDRelationship
 from .base import DataRow
 from pydantic import BaseModel, ValidationError
-from .base import subclass_registry, _SubclassRegistryBase
 from pydantic_core import PydanticUndefined
 from contextlib import suppress
 
@@ -67,68 +66,7 @@
 
 
 class NDLabel(BaseModel):
-    annotations: List[_SubclassRegistryBase]
-
-    def __init__(self, **kwargs):
-        # NOTE: Deserialization of subclasses in pydantic is difficult, see here https://blog.devgenius.io/deserialize-child-classes-with-pydantic-that-gonna-work-784230e1cf83
-        # Below implements the subclass registry as mentioned in the article. The python dicts we pass in can be missing certain fields
-        # we essentially have to infer the type against all sub classes that have the _SubclasssRegistryBase inheritance.
-        # It works by checking if the keys of our annotations we are missing in matches any required subclass.
-        # More keys are prioritized over less keys (closer match). This is used when importing json to our base models not a lot of customer workflows
-        # depend on this method but this works for all our existing tests with the bonus of added validation. (no subclass found it throws an error)
-
-        for index, annotation in enumerate(kwargs["annotations"]):
-            if isinstance(annotation, dict):
-                item_annotation_keys = annotation.keys()
-                key_subclass_combos = defaultdict(list)
-                for subclass in subclass_registry.values():
-                    # Get all required keys from subclass
-                    annotation_keys = []
-                    for k, field in subclass.model_fields.items():
-                        if field.default == PydanticUndefined and k != "uuid":
-                            if (
-                                hasattr(field, "alias")
-                                and field.alias in item_annotation_keys
-                            ):
-                                annotation_keys.append(field.alias)
-                            elif (
-                                hasattr(field, "validation_alias")
-                                and field.validation_alias
-                                in item_annotation_keys
-                            ):
-                                annotation_keys.append(field.validation_alias)
-                            else:
-                                annotation_keys.append(k)
-
-                    key_subclass_combos[subclass].extend(annotation_keys)
-
-                # Sort by subclass that has the most keys i.e. the one with the most keys that matches is most likely our subclass
-                key_subclass_combos = dict(
-                    sorted(
-                        key_subclass_combos.items(),
-                        key=lambda x: len(x[1]),
-                        reverse=True,
-                    )
-                )
-
-                for subclass, key_subclass_combo in key_subclass_combos.items():
-                    # Choose the keys from our dict we supplied that matches the required keys of a subclass
-                    check_required_keys = all(
-                        key in list(item_annotation_keys)
-                        for key in key_subclass_combo
-                    )
-                    if check_required_keys:
-                        # Keep trying subclasses until we find one that has valid values (does not throw an validation error)
-                        with suppress(ValidationError):
-                            annotation = subclass(**annotation)
-                            break
-                if isinstance(annotation, dict):
-                    raise ValueError(
-                        f"Could not find subclass for fields: {item_annotation_keys}"
-                    )
-
-            kwargs["annotations"][index] = annotation
-        super().__init__(**kwargs)
+    annotations: AnnotationType
 
     class _Relationship(BaseModel):
         """This object holds information about the relationship"""

libs/labelbox/src/labelbox/data/serialization/ndjson/metric.py

@@ -15,7 +15,6 @@
     ConfusionMatrixMetricConfidenceValue,
 )
 from pydantic import ConfigDict, model_serializer
-from .base import _SubclassRegistryBase
 
 
 class BaseNDMetric(NDJsonBase):
@@ -33,7 +32,7 @@ def serialize_model(self, handler):
         return res
 
 
-class NDConfusionMatrixMetric(BaseNDMetric, _SubclassRegistryBase):
+class NDConfusionMatrixMetric(BaseNDMetric):
     metric_value: Union[
         ConfusionMatrixMetricValue, ConfusionMatrixMetricConfidenceValue
     ]
@@ -65,7 +64,7 @@ def from_common(
         )
 
 
-class NDScalarMetric(BaseNDMetric, _SubclassRegistryBase):
+class NDScalarMetric(BaseNDMetric):
     metric_value: Union[ScalarMetricValue, ScalarMetricConfidenceValue]
     metric_name: Optional[str] = None
     aggregation: Optional[ScalarMetricAggregation] = (

libs/labelbox/src/labelbox/data/serialization/ndjson/mmc.py

@@ -2,7 +2,7 @@
 
 from labelbox.utils import _CamelCaseMixin
 
-from .base import _SubclassRegistryBase, DataRow, NDAnnotation
+from .base import DataRow, NDAnnotation
 from ...annotation_types.mmc import (
     MessageSingleSelectionTask,
     MessageMultiSelectionTask,
@@ -20,7 +20,7 @@ class MessageTaskData(_CamelCaseMixin):
     ]
 
 
-class NDMessageTask(NDAnnotation, _SubclassRegistryBase):
+class NDMessageTask(NDAnnotation):
     message_evaluation_task: MessageTaskData
 
     def to_common(self) -> MessageEvaluationTaskAnnotation: