create app from automl (#803)

Author: pplonski

supervised/apps/metadata.py

@@ -74,8 +74,7 @@ def collect_app_metadata(automl, title=None, selected_models=None):
 
 
 def _default_title(automl):
-    task = automl._ml_task.replace("_", " ").title() if automl._ml_task else "AutoML"
-    return f"MLJAR AutoML {task} App"
+    return "MLJAR AutoML"
 
 
 def _load_training_frame(automl):

supervised/apps/templates.py

@@ -154,6 +154,44 @@ def csv_download_payload(df):
             return base64.b64encode(csv_data.encode("utf-8")).decode("ascii")
 
 
+        def render_batch_dashboard(mr, scored_df):
+            task = manifest()["model_task"]
+            indicators = [mr.Indicator(value=str(len(scored_df)), label="Scored rows")]
+            if task == "regression":
+                prediction_series = pd.to_numeric(scored_df["prediction"], errors="coerce").dropna()
+                if not prediction_series.empty:
+                    indicators.append(
+                        mr.Indicator(
+                            value=f"{float(prediction_series.mean()):.6g}",
+                            label="Mean prediction",
+                        )
+                    )
+                    indicators.append(
+                        mr.Indicator(
+                            value=f"{float(prediction_series.median()):.6g}",
+                            label="Median prediction",
+                        )
+                    )
+            else:
+                label_counts = scored_df["label"].astype(str).value_counts()
+                if not label_counts.empty:
+                    top_label = str(label_counts.index[0])
+                    top_share = float(label_counts.iloc[0]) / float(len(scored_df))
+                    indicators.append(
+                        mr.Indicator(
+                            value=top_label,
+                            label="Most common label",
+                        )
+                    )
+                    indicators.append(
+                        mr.Indicator(
+                            value=f"{top_share:.1%}",
+                            label="Top label share",
+                        )
+                    )
+            _ = mr.Indicator(indicators, display_now=True)
+
+
         def render_single_dashboard(mr, result):
             if result["task"] == "regression":
                 _ = mr.Indicator(
@@ -468,7 +506,12 @@ def batch_notebook_source():
                     import mercury as mr
                     APP_IMPORT_ERROR = None
                     try:
-                        from app_support import batch_predict, csv_download_payload, plot_batch_summary
+                        from app_support import (
+                            batch_predict,
+                            csv_download_payload,
+                            plot_batch_summary,
+                            render_batch_dashboard,
+                        )
                     except Exception as exc:
                         APP_IMPORT_ERROR = f"{type(exc).__name__}: {exc}"
                         _ = mr.Markdown(
@@ -493,8 +536,10 @@ def batch_notebook_source():
                     if error_message:
                         _ = mr.Markdown(error_message)
                     elif scored_df is not None:
-                        _ = mr.Markdown(f"## Scored rows\\n\\n`{len(scored_df)}`")
-                        _ = mr.Table(scored_df.head(20))
+                        _ = mr.Markdown(
+                            "## Batch prediction results\\n\\n"
+                            "Review the scored preview below and download the full predictions CSV."
+                        )
                         mr.Download(
                             csv_download_payload(scored_df),
                             filename="predictions.csv",
@@ -503,12 +548,35 @@ def batch_notebook_source():
                             label="Download predictions",
                             position="inline",
                         )
-                        plot_batch_summary(scored_df)
                     else:
                         _ = mr.Markdown("Upload a CSV file to begin batch prediction.")
                     """
                 ).strip()
             ),
+            code_cell(
+                dedent(
+                    """
+                    if scored_df is not None:
+                        render_batch_dashboard(mr, scored_df)
+                    """
+                ).strip()
+            ),
+            code_cell(
+                dedent(
+                    """
+                    if scored_df is not None:
+                        _ = mr.Table(scored_df.head(20))
+                    """
+                ).strip()
+            ),
+            code_cell(
+                dedent(
+                    """
+                    if scored_df is not None:
+                        plot_batch_summary(scored_df)
+                    """
+                ).strip()
+            ),
         ],
         "metadata": notebook_metadata("Batch Prediction"),
         "nbformat": 4,

tests/tests_automl/test_automl_app.py

@@ -62,6 +62,7 @@ def test_app_creates_default_workspace(self):
             manifest = json.load(fin)
 
         self.assertEqual(manifest["bundle_type"], "automl_prediction_bundle")
+        self.assertEqual(manifest["title"], "MLJAR AutoML")
         self.assertEqual(manifest["default_notebook"], "predict_single.ipynb")
         self.assertEqual(len(manifest["notebooks"]), 2)
         self.assertEqual(manifest["model_task"], "multiclass_classification")