Merge pull request #21 from neptune-ai/ss/handle_NeptuneUnsupportedTy…

…pe_error

Supressing `NeptuneUnsupportedType` warning if expected metadata not found

CHANGELOG.md

@@ -1,6 +1,7 @@
 ## [UNRELEASED] neptune-sklearn 2.1.1
 
 ### Fixes
+- `create_*_summary()` now does not throw a `NeptuneUnsupportedType` error if expected metadata is not found ([#21](https://github.com/neptune-ai/neptune-sklearn/pull/21))
 - Fixed method names in docstrings ([#18](https://github.com/neptune-ai/neptune-sklearn/pull/18))
 
 ## neptune-sklearn 2.1.0

pyproject.toml

@@ -45,7 +45,7 @@ name = "neptune-sklearn"
 readme = "README.md"
 version = "0.0.0"
 classifiers = [
-    "Development Status :: 4 - Beta",
+    "Development Status :: 5 - Production/Stable",
     "Environment :: Console",
     "Intended Audience :: Developers",
     "Intended Audience :: Science/Research",

src/neptune_sklearn/impl/__init__.py

@@ -110,8 +110,8 @@ def create_regressor_summary(regressor, X_train, X_test, y_train, y_test, nrows=
         log_charts (`bool`, optional): Whether to calculate and log chart visualizations.
             Note: Calculating visualizations is potentially expensive depending on input data and regressor,
             and may take some time to finish. This is equivalent to calling the following functions from
-            this module: `create_learning_curve_chart()`, `create_feature_importance_chart()`, `create_residuals_chart()`,
-            `create_prediction_error_chart()`, and `create_cooks_distance_chart()`.
+            this module: `create_learning_curve_chart()`, `create_feature_importance_chart()`,
+            `create_residuals_chart()`, `create_prediction_error_chart()`, and `create_cooks_distance_chart()`.
 
     Returns:
         `dict` with all summary items.
@@ -144,17 +144,26 @@ def create_regressor_summary(regressor, X_train, X_test, y_train, y_test, nrows=
         "scores": get_scores(regressor, X_test, y_test, y_pred=y_pred),
     }
 
-    if log_charts:
-        reg_summary["diagnostics_charts"] = {
-            "learning_curve": create_learning_curve_chart(regressor, X_train, y_train),
-            "feature_importance": create_feature_importance_chart(regressor, X_train, y_train),
-            "residuals": create_residuals_chart(regressor, X_train, X_test, y_train, y_test),
-            "prediction_error": create_prediction_error_chart(regressor, X_train, X_test, y_train, y_test),
-            "cooks_distance": create_cooks_distance_chart(regressor, X_train, y_train),
-        }
-
     reg_summary["integration/about/neptune-sklearn"] = __version__
 
+    if log_charts:
+        learning_curve = create_learning_curve_chart(regressor, X_train, y_train)
+        feature_importance = create_feature_importance_chart(regressor, X_train, y_train)
+        residuals = create_residuals_chart(regressor, X_train, X_test, y_train, y_test)
+        prediction_error = create_prediction_error_chart(regressor, X_train, X_test, y_train, y_test)
+        cooks_distance = create_cooks_distance_chart(regressor, X_train, y_train)
+
+        if learning_curve:
+            reg_summary["diagnostics_charts/learning_curve"] = learning_curve
+        if feature_importance:
+            reg_summary["diagnostics_charts/feature_importance"] = feature_importance
+        if residuals:
+            reg_summary["diagnostics_charts/residuals"] = residuals
+        if prediction_error:
+            reg_summary["diagnostics_charts/prediction_error"] = prediction_error
+        if cooks_distance:
+            reg_summary["diagnostics_charts/cooks_distance"] = cooks_distance
+
     return reg_summary
 
 
@@ -217,17 +226,26 @@ def create_classifier_summary(classifier, X_train, X_test, y_train, y_test, nrow
         "scores": get_scores(classifier, X_test, y_test, y_pred=y_pred),
     }
 
-    if log_charts:
-        cls_summary["diagnostics_charts"] = {
-            "classification_report": create_classification_report_chart(classifier, X_train, X_test, y_train, y_test),
-            "confusion_matrix": create_confusion_matrix_chart(classifier, X_train, X_test, y_train, y_test),
-            "ROC_AUC": create_roc_auc_chart(classifier, X_train, X_test, y_train, y_test),
-            "precision_recall": create_precision_recall_chart(classifier, X_test, y_test),
-            "class_prediction_error": create_class_prediction_error_chart(classifier, X_train, X_test, y_train, y_test),
-        }
-
     cls_summary["integration/about/neptune-sklearn"] = __version__
 
+    if log_charts:
+        classification_report = create_classification_report_chart(classifier, X_train, X_test, y_train, y_test)
+        confusion_matrix = create_confusion_matrix_chart(classifier, X_train, X_test, y_train, y_test)
+        roc_auc = create_roc_auc_chart(classifier, X_train, X_test, y_train, y_test)
+        precision_recall = create_precision_recall_chart(classifier, X_test, y_test)
+        class_prediction_error = create_class_prediction_error_chart(classifier, X_train, X_test, y_train, y_test)
+
+        if classification_report:
+            cls_summary["diagnostics_charts/classification_report"] = classification_report
+        if confusion_matrix:
+            cls_summary["diagnostics_charts/confusion_matrix"] = confusion_matrix
+        if roc_auc:
+            cls_summary["diagnostics_charts/ROC_AUC"] = roc_auc
+        if precision_recall:
+            cls_summary["diagnostics_charts/precision_recall"] = precision_recall
+        if class_prediction_error:
+            cls_summary["diagnostics_charts/class_prediction_error"] = class_prediction_error
+
     return cls_summary
 
 
@@ -272,13 +290,16 @@ def create_kmeans_summary(model, X, nrows=1000, **kwargs):
     kmeans_summary["all_params"] = stringify_unsupported(get_estimator_params(model))
     kmeans_summary["pickled_model"] = get_pickled_model(model)
     kmeans_summary["cluster_labels"] = get_cluster_labels(model, X, nrows=nrows, **kwargs)
-    kmeans_summary["diagnostics_charts"] = {
-        "kelbow": create_kelbow_chart(model, X, **kwargs),
-        "silhouette": create_silhouette_chart(model, X, **kwargs),
-    }
-
     kmeans_summary["integration/about/neptune-sklearn"] = __version__
 
+    kelbow = create_kelbow_chart(model, X, **kwargs)
+    silhouette = create_silhouette_chart(model, X, **kwargs)
+
+    if kelbow:
+        kmeans_summary["diagnostics_charts/kelbow"] = kelbow
+    if silhouette:
+        kmeans_summary["diagnostics_charts/silhouette"] = silhouette
+
     return kmeans_summary
 
 

tests/test_e2e.py

@@ -3,13 +3,18 @@
 except ImportError:
     from neptune.new import init_run
 
+import pytest
 from sklearn import datasets
 from sklearn.cluster import KMeans
+from sklearn.dummy import DummyRegressor
 from sklearn.linear_model import (
     LinearRegression,
     LogisticRegression,
 )
-from sklearn.model_selection import train_test_split
+from sklearn.model_selection import (
+    GridSearchCV,
+    train_test_split,
+)
 
 import neptune_sklearn as npt_utils
 
@@ -61,6 +66,33 @@ def test_kmeans_summary():
     validate_run(run, log_charts=True)
 
 
+@pytest.mark.filterwarnings("error::neptune.common.warnings.NeptuneUnsupportedType")
+def test_unsupported_object():
+    """This method checks if Neptune throws a `NeptuneUnsupportedType` warning if expected metadata
+    is not found or skips trying to log such metadata"""
+
+    with init_run() as run:
+
+        X, y = datasets.load_diabetes(return_X_y=True)
+        X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.5)
+
+        model = DummyRegressor()
+
+        param_grid = {
+            "strategy": ["mean", "median", "quantile"],
+            "quantile": [0.1, 0.5, 1.0],
+        }
+
+        X, y = datasets.fetch_california_housing(return_X_y=True)[:10]
+        X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2)
+
+        grid_cv = GridSearchCV(model, param_grid, scoring="neg_mean_absolute_error", cv=2).fit(X_train, y_train)
+
+        run["regressor_summary"] = npt_utils.create_regressor_summary(grid_cv, X_train, X_test, y_train, y_test)
+
+        run.wait()
+
+
 def validate_run(run, log_charts):
     assert run.exists("summary/all_params")
     assert run.exists("summary/pickled_model")