Completes OPEN-5006 Infer config properties from dataset/model

Author: gustavocidornelas

openlayer/__init__.py

@@ -718,6 +718,7 @@ def add_dataset(
                 "Either `dataset_config` or `dataset_config_file_path` must be"
                 " provided."
             )
+
         # Validate dataset
         dataset_validator = dataset_validators.get_validator(
             task_type=task_type,
@@ -739,6 +740,8 @@ def add_dataset(
         dataset_data = DatasetSchema().load(
             {"task_type": task_type.value, **dataset_config}
         )
+        if dataset_data.get("columnNames") is None:
+            dataset_data["columnNames"] = utils.get_column_names(file_path)
 
         # Copy relevant resources to temp directory
         with tempfile.TemporaryDirectory() as temp_dir:

openlayer/schemas.py

@@ -76,7 +76,8 @@ class BaseDatasetSchema(ma.Schema):
 
     columnNames = ma.fields.List(
         ma.fields.Str(validate=COLUMN_NAME_VALIDATION_LIST),
-        required=True,
+        allow_none=True,
+        load_default=None,
     )
     label = ma.fields.Str(
         validate=ma.validate.OneOf(
@@ -233,11 +234,12 @@ class BaseModelSchema(ma.Schema):
     """Common schema for models for all task types."""
 
     name = ma.fields.Str(
-        required=True,
         validate=ma.validate.Length(
             min=1,
             max=64,
         ),
+        allow_none=True,
+        load_default="Model",
     )
     metadata = ma.fields.Dict(
         allow_none=True,
@@ -252,7 +254,8 @@ class BaseModelSchema(ma.Schema):
             + " If you can't find your framework, specify 'custom' for your model's"
             + " `architectureType`.",
         ),
-        required=True,
+        allow_none=True,
+        load_default="custom",
     )
 
 
@@ -307,6 +310,19 @@ class LLMModelSchema(BaseModelSchema):
         ma.fields.Str(validate=COLUMN_NAME_VALIDATION_LIST),
         load_default=[],
     )
+    # Important that here the architectureType defaults to `llm` and not `custom` since
+    # the architectureType is used to check if the model is an LLM or not.
+    architectureType = ma.fields.Str(
+        validate=ma.validate.OneOf(
+            [model_framework.value for model_framework in ModelType],
+            error="`architectureType` must be one of the supported frameworks."
+            + " Check out our API reference for a full list."
+            + " If you can't find your framework, specify 'custom' for your model's"
+            + " `architectureType`.",
+        ),
+        allow_none=True,
+        load_default="llm",
+    )
 
     @ma.validates_schema
     def validates_model_type_fields(self, data, **kwargs):

openlayer/utils.py

@@ -77,6 +77,18 @@ def camel_to_snake_str(name: str) -> str:
     return re.sub(r"(?<!^)(?=[A-Z])", "_", name).lower()
 
 
+def get_column_names(file_path: str) -> list:
+    """Returns the column names of the specified file.
+
+    Args:
+        file_path (str): the path to the file.
+
+    Returns:
+        list: the column names of the specified file.
+    """
+    return pd.read_csv(file_path, nrows=0).columns.tolist()
+
+
 def get_exception_stacktrace(err: Exception):
     """Returns the stacktrace of the most recent exception.
 

openlayer/validators/dataset_validators.py

@@ -162,7 +162,6 @@ def _validate_dataset_and_config_consistency(self):
 
             # Dataset-wide validations
             self._validate_dataset_dtypes()
-            self._validate_dataset_columns()
 
             self._validate_inputs()
             self._validate_outputs()
@@ -180,16 +179,6 @@ def _validate_dataset_dtypes(self):
                 " Please cast the columns in your dataset to conform to these dtypes."
             )
 
-    def _validate_dataset_columns(self):
-        """Checks whether all columns in the dataset are specified in `columnNames`."""
-        dataset_columns = set(self.dataset_df.columns)
-        column_names = set(self.column_names)
-        if dataset_columns != column_names:
-            self.failed_validations.append(
-                "Not all columns in the dataset are specified in `columnNames`. "
-                "Please specify all dataset columns in `columnNames`."
-            )
-
     @abstractmethod
     def _validate_inputs(self):
         """To be implemented by InputValidator child classes."""