fix: RND-101: Fix ML backend compatibility with sdk>=1 (#570)

niklub · nik · web-flow · commit bdfc42b68a0a · 2024-06-19T14:01:52.000+01:00
* fix: RND-101: Fix ML backend compatibility with sdk&gt;=1

* Fix sdk version

* Install sdk from git

* Fix label-studio-tools imports

* Install numpy in requirements

* Cast float

* Add numpy reqs to easyocr and nemo

* Upgrade label-studio-sdk==1.0.2

---------

Co-authored-by: nik &lt;nik@heartex.net&gt;
diff --git a/label_studio_ml/api.py b/label_studio_ml/api.py
@@ -77,7 +77,7 @@ def _predict():
         else:
             response.update_predictions_version()
 
-        response = response.serialize()
+        response = response.model_dump()
 
     res = response
     if res is None:
diff --git a/label_studio_ml/examples/easyocr/requirements.txt b/label_studio_ml/examples/easyocr/requirements.txt
@@ -1,3 +1,4 @@
 easyocr==1.7.1
 boto3==1.28.58
-opencv-python-headless==4.9.0.80
+opencv-python-headless==4.9.0.80
+numpy<2
diff --git a/label_studio_ml/examples/huggingface_ner/model.py b/label_studio_ml/examples/huggingface_ner/model.py
@@ -83,7 +83,7 @@ def predict(self, tasks: List[Dict], context: Optional[Dict] = None, **kwargs) -
                 entities = list(group)
                 start = entities[0]['start']
                 end = entities[-1]['end']
-                score = sum([entity['score'] for entity in entities]) / len(entities)
+                score = float(sum([entity['score'] for entity in entities]) / len(entities))
                 results.append({
                     'from_name': from_name,
                     'to_name': to_name,
diff --git a/label_studio_ml/examples/mmdetection-3/requirements.txt b/label_studio_ml/examples/mmdetection-3/requirements.txt
@@ -1,2 +1,3 @@
 boto3>=1.26.103,<2.0.0
 openmim~=0.3.9
+numpy~=1.26
diff --git a/label_studio_ml/examples/nemo_asr/requirements.txt b/label_studio_ml/examples/nemo_asr/requirements.txt
@@ -1 +1,2 @@
 nemo-toolkit[all]
+numpy<2
diff --git a/label_studio_ml/examples/spacy/requirements.txt b/label_studio_ml/examples/spacy/requirements.txt
@@ -1,3 +1,4 @@
 gunicorn==22.0.0
 spacy~=3.6
 label-studio-ml @ git+https://github.com/HumanSignal/label-studio-ml-backend.git@master
+numpy~=1.26
diff --git a/label_studio_ml/model.py b/label_studio_ml/model.py
@@ -24,8 +24,8 @@
 from colorama import Fore
 
 from label_studio_sdk.label_interface import LabelInterface
-from label_studio_tools.core.label_config import parse_config
-from label_studio_tools.core.utils.io import get_local_path
+from label_studio_sdk._extensions.label_studio_tools.core.label_config import parse_config
+from label_studio_sdk._extensions.label_studio_tools.core.utils.io import get_local_path
 from .response import ModelResponse
 from .utils import is_preload_needed
 from .cache import create_cache
diff --git a/label_studio_ml/response.py b/label_studio_ml/response.py
@@ -1,15 +1,19 @@
 
 from typing import Type, Dict, Optional, List, Tuple, Any, Union
-from pydantic import BaseModel, confloat
+from pydantic import BaseModel, confloat, Field
+from label_studio_sdk.label_interface.objects import PredictionValue
+from typing import Union, List
 
-from label_studio_sdk.objects import PredictionValue
+
+# one or multiple predictions per task
+SingleTaskPredictions = Union[List[PredictionValue], PredictionValue]
 
 
 class ModelResponse(BaseModel):
     """
     """
     model_version: Optional[str] = None
-    predictions: List[PredictionValue]
+    predictions: List[SingleTaskPredictions]
 
     def has_model_version(self) -> bool:
         return bool(self.model_version)
@@ -18,21 +22,16 @@ def update_predictions_version(self) -> None:
         """
         """
         for prediction in self.predictions:
-            if not prediction.model_version:
-                prediction.model_version = self.model_version
+            if isinstance(prediction, PredictionValue):
+                prediction = [prediction]
+            for p in prediction:
+                if not p.model_version:
+                    p.model_version = self.model_version
     
     def set_version(self, version: str) -> None:
         """
         """
         self.model_version = version
         # Set the version for each prediction
         self.update_predictions_version()
-
-    def serialize(self):
-        """
-        """
-        return {
-            "model_version": self.model_version,
-            "predictions": [ p.serialize() for p in self.predictions ]
-        }
         
diff --git a/label_studio_ml/utils.py b/label_studio_ml/utils.py
@@ -8,9 +8,8 @@
 from typing import List
 from urllib.parse import urlparse
 
-from label_studio_tools.core.utils.params import get_env
-from label_studio_tools.core.utils.io import get_local_path
-from label_studio_sdk.label_interface import LabelInterface
+from label_studio_sdk._extensions.label_studio_tools.core.utils.params import get_env
+from label_studio_sdk._extensions.label_studio_tools.core.utils.io import get_local_path
 
 DATA_UNDEFINED_NAME = '$undefined$'
 
diff --git a/requirements.txt b/requirements.txt
@@ -3,5 +3,5 @@ colorama~=0.4
 requests~=2.31
 semver~=3.0.2
 pillow~=10.3
-label_studio_tools @ git+https://github.com/HumanSignal/label-studio-tools.git@master
-label-studio-sdk==0.0.34
+label-studio-sdk==1.0.2
+# label-studio-sdk @ git+https://github.com/HumanSignal/label-studio-sdk.git
diff --git a/tests/test_response.py b/tests/test_response.py
@@ -4,15 +4,21 @@
 from typing import Optional
 
 from label_studio_ml.response import ModelResponse
-from label_studio_sdk.objects import PredictionValue
+from label_studio_sdk.label_interface.objects import PredictionValue
 
-MODEL_VERSION = 0.1
+MODEL_VERSION = "0.1"
 
 @pytest.fixture
 def predictions():
     # Replace 'Example Prediction' with real Predictions
-    return [PredictionValue(model_version=MODEL_VERSION, score=0.1, result=[{}]), 
-            PredictionValue(score=0.1, result=[{}])] 
+    return [
+        PredictionValue(model_version=MODEL_VERSION, score=0.1, result=[{}]),
+        PredictionValue(score=0.1, result=[{}]),
+        [
+            PredictionValue(model_version='prediction_1', score=0.2, result=[{}]),
+            PredictionValue(score=0.3, result=[{}]),
+        ]
+    ]
 
 def test_has_model_version(predictions):
     res = ModelResponse(predictions=predictions)
@@ -45,8 +51,27 @@ def test_set_version(predictions):
 def test_serialize(predictions):
     # Assuming PredictionValue has method .serialize() which returns a dict 
     res = ModelResponse(model_version="1.0", predictions=predictions)
-    serialized_res = res.serialize()
+    serialized_res = res.model_dump()
 
     assert serialized_res['model_version'] == "1.0"
     assert serialized_res['predictions'] == [
-        p.serialize() for p in predictions] 
+        {'model_version': MODEL_VERSION, 'score': 0.1, 'result': [{}]},
+        {'model_version': None, 'score': 0.1, 'result': [{}]},
+        [
+            {'model_version': 'prediction_1', 'score': 0.2, 'result': [{}]},
+            {'model_version': None, 'score': 0.3, 'result': [{}]},
+        ]
+    ]
+
+    # update model_version
+    res.update_predictions_version()
+    serialized_res = res.model_dump()
+
+    assert serialized_res['predictions'] == [
+        {'model_version': "0.1", 'score': 0.1, 'result': [{}]},
+        {'model_version': "1.0", 'score': 0.1, 'result': [{}]},
+        [
+            {'model_version': 'prediction_1', 'score': 0.2, 'result': [{}]},
+            {'model_version': '1.0', 'score': 0.3, 'result': [{}]},
+        ]
+    ]

Original file line number	Diff line number	Diff line change
`@@ -1,2 +1,3 @@`
`1`	`1`	`boto3>=1.26.103,<2.0.0`
`2`	`2`	`openmim~=0.3.9`
	`3`	`+numpy~=1.26`