[GSK-1956] Moved target validation and silenced it for LLMs

giskard/core/dataset_validation.py

@@ -8,18 +8,20 @@
 from giskard.datasets.base import Dataset
 
 
-def validate_target(ds: Dataset):
+def validate_optional_target(ds: Dataset):
     if not ds.target:
         warning(
             "You did not provide the optional argument 'target'. "
             "'target' is the column name in df corresponding to the actual target variable (ground truth)."
         )
-    else:
-        if ds.target not in ds.columns:
-            raise ValueError(
-                "Invalid target parameter:"
-                f" '{ds.target}' column is not present in the dataset with columns: {list(ds.columns)}"
-            )
+
+
+def validate_target_exists(ds: Dataset):
+    if ds.target and ds.target not in ds.columns:
+        raise ValueError(
+            "Invalid target parameter:"
+            f" '{ds.target}' column is not present in the dataset with columns: {list(ds.columns)}"
+        )
 
 
 def validate_dtypes(ds: Dataset):

giskard/core/model_validation.py

@@ -14,11 +14,14 @@
 from giskard.models.base import BaseModel, WrapperModel
 from ..utils import fullname
 from ..utils.analytics_collector import analytics, get_dataset_properties, get_model_properties
+from .dataset_validation import validate_optional_target
 
 
 @configured_validate_arguments
 def validate_model(model: BaseModel, validate_ds: Optional[Dataset] = None, print_validation_message: bool = True):
     try:
+        if model.meta.model_type != SupportedModelTypes.TEXT_GENERATION and validate_ds is not None:
+            validate_optional_target(validate_ds)
         _do_validate_model(model, validate_ds)
     except (ValueError, TypeError) as err:
         _track_validation_error(err, model, validate_ds)

giskard/datasets/base/__init__.py

@@ -186,10 +186,10 @@ def __init__(
         self.target = target
 
         if validation:
-            from giskard.core.dataset_validation import validate_dtypes, validate_target
+            from giskard.core.dataset_validation import validate_dtypes, validate_target_exists
 
             validate_dtypes(self)
-            validate_target(self)
+            validate_target_exists(self)
 
         self.column_dtypes = self.extract_column_dtypes(self.df)
 

tests/test_dataset.py

@@ -7,6 +7,7 @@
 import requests_mock
 
 from giskard.datasets.base import Dataset
+from giskard.core.dataset_validation import validate_optional_target
 from giskard.client.dtos import DatasetMetaInfo
 
 from tests import utils
@@ -29,7 +30,6 @@
     "target",
 ]
 
-
 valid_df = pd.DataFrame(
     {
         "categorical_column": ["turtle", "crocodile", "turtle"],
@@ -65,6 +65,7 @@ def test_valid_df_column_types():
         r"in df corresponding to the actual target variable \(ground truth\)\.",
     ):
         my_dataset = Dataset(valid_df)
+        validate_optional_target(my_dataset)
     assert my_dataset.column_types == {
         "categorical_column": "category",
         "text_column": "text",