Fix warnings emitted by the application scorecard full suite

pyproject.toml

@@ -27,7 +27,7 @@ dependencies = [
   "plotly (>=6.0.0)",
   "polars",
   "python-dotenv",
-  "scikit-learn",
+  "scikit-learn (>=1.4)",
   "seaborn",
   "tabulate (>=0.9.0,<0.10.0)",
   "tiktoken",
@@ -41,7 +41,7 @@ dependencies = [
 all = [
   "torch (>=2.0.0)",
   "xgboost (>=1.5.2,<3.1)",
-  "scikit-learn (<1.8)",
+  "scikit-learn (>=1.4,<1.8)",
   "transformers (>=4.32.0,<5.0.0)",
   "pycocoevalcap",
   "ragas (>=0.2.3,<=0.2.7)",
@@ -111,7 +111,7 @@ pytorch = ["torch (>=2.0.0)"]
 stats = ["scipy", "statsmodels (>=0.14.2,<0.15.0)", "arch (>=7.0.0)"]
 xgboost = [
   "xgboost (>=1.5.2,<3.1)",
-  "scikit-learn (<1.8)",
+  "scikit-learn (>=1.4,<1.8)",
 ]
 explainability = [
   "shap (>=0.46.0)",

tests/test_test_descriptions.py

@@ -17,6 +17,29 @@
 class TestTokenEstimation(unittest.TestCase):
     """Test token estimation and truncation functions."""
 
+    @patch("validmind.api_client.generate_test_result_description")
+    def test_generate_description_params_are_json_serializable(self, mock_generate):
+        """Params holding DataFrames (e.g. fit_params eval_set) must not break the request"""
+        import json
+
+        import pandas as pd
+
+        from validmind.ai.test_descriptions import generate_description
+        from validmind.vm_models.result import ResultTable
+
+        mock_generate.return_value = {"content": "ok"}
+
+        generate_description(
+            test_id="validmind.model_validation.sklearn.HyperParametersTuning",
+            test_description="desc",
+            tables=[ResultTable(data=[{"Optimized for": "recall", "recall": 0.9}])],
+            params={"eval_set": [(pd.DataFrame({"a": [1]}), pd.Series([0]))]},
+        )
+
+        payload = mock_generate.call_args[0][0]
+        json.dumps(payload)  # stdlib encoder, as used by requests
+        self.assertIn("eval_set", payload["params"])
+
     def test_estimate_tokens_simple(self):
         """Test simple character-based token estimation."""
         # Test with empty string

validmind/ai/test_descriptions.py

@@ -171,7 +171,9 @@ def generate_description(
             "figures": [figure._get_b64_url() for figure in figures or []],
             "additional_context": additional_context,
             "instructions": instructions,
-            "params": params,
+            # params can hold anything the test accepts (DataFrames, arrays);
+            # round-trip so the request body is plain JSON
+            "params": json.loads(json.dumps(params, cls=NumpyEncoder)),
         }
     )["content"]
 

validmind/datasets/credit_risk/lending_club.py

@@ -322,7 +322,9 @@ def woe_encoding(df: pd.DataFrame, verbose: bool = True) -> pd.DataFrame:
             print(f"Excluded {target_column} from WoE transformation.")
 
     # Apply the WoE transformation
-    df = sc.woebin_ply(df, bins=bins)
+    with warnings.catch_warnings():
+        warnings.simplefilter("ignore")  # scorecardpy uses deprecated pandas APIs
+        df = sc.woebin_ply(df, bins=bins)
 
     if verbose:
         print("Successfully converted features to WoE values.")
@@ -379,7 +381,9 @@ def _woebin(df: pd.DataFrame, verbose: bool = True) -> Dict[str, Any]:
             print(
                 f"Performing binning with breaks_adj: {breaks_adj}"
             )  # print the breaks_adj being used
-        bins = sc.woebin(df, target_column, breaks_list=breaks_adj)
+        with warnings.catch_warnings():
+            warnings.simplefilter("ignore")  # scorecardpy uses deprecated pandas APIs
+            bins = sc.woebin(df, target_column, breaks_list=breaks_adj)
     except Exception as e:
         print("Error during binning: ")
         print(e)

validmind/tests/model_validation/sklearn/HyperParametersTuning.py

@@ -43,7 +43,7 @@ def _create_scoring_dict(scoring, metrics, threshold):
     for metric in metrics:
         if metric == "recall":
             scoring_dict[metric] = make_scorer(
-                custom_recall, needs_proba=True, threshold=threshold
+                custom_recall, response_method="predict_proba", threshold=threshold
             )
         elif metric == "roc_auc":
             scoring_dict[metric] = "roc_auc"

validmind/tests/model_validation/sklearn/ScoreProbabilityAlignment.py

@@ -81,7 +81,7 @@ def ScoreProbabilityAlignment(
 
     # Calculate statistics per bin
     results = []
-    for bin_name, group in df.groupby("score_bin"):
+    for bin_name, group in df.groupby("score_bin", observed=True):
         bin_stats = {
             "Score Range": f"{bin_name.left:.0f}-{bin_name.right:.0f}",
             "Mean Score": group[score_column].mean(),

validmind/tests/model_validation/sklearn/WeakspotsDiagnosis.py

@@ -285,14 +285,22 @@ def WeakspotsDiagnosis(
     figures = []
     passed = True
 
-    df_1 = datasets[0]._df[
-        feature_columns
-        + [datasets[0].target_column, datasets[0].prediction_column(model)]
-    ]
-    df_2 = datasets[1]._df[
-        feature_columns
-        + [datasets[1].target_column, datasets[1].prediction_column(model)]
-    ]
+    df_1 = (
+        datasets[0]
+        ._df[
+            feature_columns
+            + [datasets[0].target_column, datasets[0].prediction_column(model)]
+        ]
+        .copy()
+    )
+    df_2 = (
+        datasets[1]
+        ._df[
+            feature_columns
+            + [datasets[1].target_column, datasets[1].prediction_column(model)]
+        ]
+        .copy()
+    )
     results_1 = pd.DataFrame()
     results_2 = pd.DataFrame()
     for feature in feature_columns: