Merge upstream/main into autox_clean_up_stages

Resolved conflict in timeseries_data_loader/component.py:
- Kept enhanced sample_rows logic with ISO timestamp conversion (from PR opendatahub-io#132)
- Kept display_name metadata at start of context (from this PR)
- Kept write_outputs status tracking (from PR opendatahub-io#132)

Merged changes:
- PR opendatahub-io#132: AutoML timeseries notebook backtesting charts
- PR opendatahub-io#138: ai4rag 0.6.4 and ogx-client 1.1.0 updates

Signed-off-by: Lukasz Cmielowski <lcmielow@redhat.com>
Assisted-by: Cursor

Author: LukaszCmielowski

components/data_processing/automl/timeseries_data_loader/component.py

@@ -61,6 +61,7 @@ def timeseries_data_loader(
         NamedTuple: sample_config, split_config, sample_rows, models_selection_train_data_path, extra_train_data_path.
     """
     import io
+    import json
     import logging
     import os
     from pathlib import Path
@@ -421,8 +422,19 @@ def _concat_sorted(parts: list, sort_by: list) -> pd.DataFrame:
             "selection_train_size": selection_train_size,
         }
 
-        # Sample row for downstream use (JSON string to avoid NaN issues)
-        sample_rows = test_df.tail(min(5, len(test_df))).to_json(orient="records")
+        status.record("write_outputs", "started")
+        status.record("write_outputs", "completed")
+
+        # Sample rows for downstream use (ISO timestamps when supported; JSON string to avoid NaN issues)
+        sample_tail = test_df.tail(min(5, len(test_df)))
+        if hasattr(sample_tail, "to_dict"):
+            from kfp_components.components.training.automl.shared.timeseries_notebook_utils import (
+                _json_records,
+            )
+
+            sample_rows = json.dumps(_json_records(sample_tail))
+        else:
+            sample_rows = sample_tail.to_json(orient="records")
 
         return NamedTuple(
             "outputs",

components/data_processing/automl/timeseries_data_loader/tests/test_component_unit.py

@@ -279,6 +279,7 @@ def test_sample_rows_json_matches_test_tail(self, tmp_path):
         assert isinstance(parsed, list)
         assert len(parsed) == 5
         assert parsed[-1]["target"] == "99"
+        assert isinstance(parsed[0]["timestamp"], str)
 
     @mock.patch.dict(os.environ, mocked_env_variables, clear=True)
     def test_sampling_truncates_below_minimum_raises(self, tmp_path):

components/data_processing/autorag/documents_discovery/README.md

@@ -63,7 +63,7 @@ def example_pipeline(
     - Name: Pipelines, Version: >=2.15.2
   - External Services:
     - Name: RHOAI Connections API, Version: >=1.0.0
-    - Name: ai4rag, Version: ~=0.6.3
+    - Name: ai4rag, Version: ~=0.6.4
 - **Tags**:
   - data-processing
   - autorag

components/data_processing/autorag/documents_discovery/metadata.yaml

@@ -9,7 +9,7 @@ dependencies:
     - name: RHOAI Connections API
       version: ">=1.0.0"
     - name: ai4rag
-      version: "~=0.6.3"
+      version: "~=0.6.4"
 tags:
   - data-processing
   - autorag

components/data_processing/autorag/documents_indexing/README.md

@@ -79,7 +79,7 @@ def example_pipeline(
     - Name: Pipelines, Version: >=2.15.2
   - External Services:
     - Name: RHOAI Connections API, Version: >=1.0.0
-    - Name: ai4rag, Version: ~=0.6.3
+    - Name: ai4rag, Version: ~=0.6.4
 - **Tags**:
   - data-indexing
   - autorag

components/data_processing/autorag/documents_indexing/metadata.yaml

@@ -9,7 +9,7 @@ dependencies:
     - name: RHOAI Connections API
       version: ">=1.0.0"
     - name: ai4rag
-      version: "~=0.6.3"
+      version: "~=0.6.4"
 tags:
   - data-indexing
   - autorag

components/training/automl/autogluon_timeseries_models_training/README.md

@@ -140,6 +140,4 @@ Under each ``{model_name}_FULL/metrics/`` directory:
 - **`back_testing.json`**: Multi-window backtest with ``per_window_metrics`` and ``series_analysis``
   (best/worst forecast timelines). Window error metrics use **natural positive** signs via
   ``filter_finite_metrics``. Best-effort after refit; omitted if backtest APIs or history are
-  insufficient.
-
-The timeseries notebook template loads ``back_testing.json`` when present for model insights.
+  insufficient. Visualized in the generated inference notebook when present.

components/training/automl/autogluon_timeseries_models_training/component.py

@@ -83,8 +83,12 @@ def autogluon_timeseries_models_training(
     logger = logging.getLogger(__name__)
 
     from kfp_components.components.training.automl.shared.back_testing import build_back_testing_json
+    from kfp_components.components.training.automl.shared.back_testing_charts import (
+        notebook_backtest_charts_source,
+    )
     from kfp_components.components.training.automl.shared.timeseries_notebook_utils import (
         build_predict_sample_artifact,
+        notebook_timeseries_sample_helpers_source,
     )
 
     status = ComponentStatusTracker(component_status.path, "autogluon_timeseries_models_training")
@@ -399,6 +403,8 @@ def replace_placeholder_in_notebook(notebook, replacements):
                     "<REPLACE_PIPELINE_NAME>": pipeline_name_trimmed,
                     "<REPLACE_MODEL_NAME>": model_name_full,
                     "<REPLACE_PREDICT_SAMPLE>": str(predict_sample),
+                    "<REPLACE_BACKTEST_PLOT_HELPERS>": notebook_backtest_charts_source(),
+                    "<REPLACE_TIMESERIES_SAMPLE_HELPERS>": notebook_timeseries_sample_helpers_source(),
                 }
                 notebook = replace_placeholder_in_notebook(notebook, replacements)
 

components/training/automl/shared/back_testing.py

@@ -140,34 +140,30 @@ def _mean_prediction_column(predictions: pd.DataFrame) -> str:
 
 
 def _quantile_bounds(predictions: pd.DataFrame) -> tuple[str | None, str | None]:
-    """Extract lower/upper quantile columns for uncertainty visualization.
+    """Extract lower/upper quantile columns aligned with ``TimeSeriesPredictor.plot`` defaults.
 
-    Selects the first quantile <= 0.2 for lower bound and >= 0.8 for upper bound.
-    These thresholds provide a ~60% coverage interval suitable for chart rendering
-    without overwhelming the visualization with too many bands.
-
-    AutoGluon typically generates quantiles at [0.1, 0.2, ..., 0.9]. If present,
-    this function will select 0.2 and 0.8. If AutoGluon uses different levels
-    (e.g., 0.1/0.9), the function adapts by picking the closest match.
-
-    Args:
-        predictions: Forecast DataFrame with quantile columns (numeric names like "0.1")
-
-    Returns:
-        Tuple of (lower_quantile_column, upper_quantile_column) as strings, or None if not found
+    Prefers quantile levels closest to 0.1 and 0.9 (AutoGluon's typical P10/P90 band).
     """
-    lower = None
-    upper = None
+    levels: list[tuple[float, str]] = []
     for col in predictions.columns:
-        col_text = str(col)
         try:
-            level = float(col)
+            levels.append((float(col), str(col)))
         except (TypeError, ValueError):
             continue
-        if level <= 0.2 and lower is None:
-            lower = col_text
-        if level >= 0.8 and upper is None:
-            upper = col_text
+    if not levels:
+        return None, None
+
+    def _closest(candidates: list[tuple[float, str]], target: float) -> str | None:
+        if not candidates:
+            return None
+        return min(candidates, key=lambda item: abs(item[0] - target))[1]
+
+    lower = _closest(levels, 0.1)
+    if lower is None:
+        return None, None
+
+    remaining = [(value, name) for value, name in levels if name != lower]
+    upper = _closest(remaining, 0.9)
     return lower, upper
 
 
@@ -244,10 +240,12 @@ def _forecast_data_for_item(
             lower = to_finite_float(pred_item.loc[ts, lower_col])
             if lower is not None:
                 row["lower_bound"] = lower
+                row["lower_quantile"] = float(lower_col)
         if upper_col is not None:
             upper = to_finite_float(pred_item.loc[ts, upper_col])
             if upper is not None:
                 row["upper_bound"] = upper
+                row["upper_quantile"] = float(upper_col)
         rows.append(row)
     return rows
 
@@ -502,6 +500,7 @@ def build_back_testing_json(
         per_window_metrics.append(
             {
                 "window_id": window_id,
+                "cutoff": cutoff,
                 "test_start": test_start,
                 "test_end": test_end,
                 "metrics": window_scores,