improve auto-select logic and handle missing data

Author: codeloop

ads/opctl/operator/lowcode/common/transformations.py

@@ -329,6 +329,8 @@ def build_fforms_meta_features(self, data, target_col=None, group_cols=None):
         if target_col not in data.columns:
             raise ValueError(f"Target column '{target_col}' not found in DataFrame")
 
+        data[target_col] = data[target_col].fillna(0)
+
         # Check if group_cols are provided and valid
         if group_cols is not None:
             if not isinstance(group_cols, list):

ads/opctl/operator/lowcode/forecast/__main__.py

@@ -9,14 +9,13 @@
 import sys
 from typing import Dict, List
 
-import pandas as pd
 import yaml
 
 from ads.opctl import logger
 from ads.opctl.operator.common.const import ENV_OPERATOR_ARGS
 from ads.opctl.operator.common.utils import _parse_input_args
 
-from .const import AUTO_SELECT_SERIES
+from .const import AUTO_SELECT, AUTO_SELECT_SERIES
 from .model.forecast_datasets import ForecastDatasets, ForecastResults
 from .operator_config import ForecastOperatorConfig
 from .whatifserve import ModelDeploymentManager
@@ -29,8 +28,10 @@ def operate(operator_config: ForecastOperatorConfig) -> ForecastResults:
     datasets = ForecastDatasets(operator_config)
     model = ForecastOperatorModelFactory.get_model(operator_config, datasets)
 
-    if operator_config.spec.model == AUTO_SELECT_SERIES and hasattr(
-        operator_config.spec, "meta_features"
+    if (
+        operator_config.spec.model == AUTO_SELECT_SERIES
+        and hasattr(operator_config.spec, "meta_features")
+        and operator_config.spec.target_category_columns
     ):
         # For AUTO_SELECT_SERIES, handle each series with its specific model
         meta_features = operator_config.spec.meta_features
@@ -64,8 +65,6 @@ def operate(operator_config: ForecastOperatorConfig) -> ForecastResults:
             )
             sub_results_list.append(sub_results)
 
-            # results_df = pd.concat([results_df, sub_result_df], ignore_index=True, axis=0)
-            # elapsed_time += sub_elapsed_time
         # Merge all sub_results into a single ForecastResults object
         if sub_results_list:
             results = sub_results_list[0]
@@ -75,6 +74,15 @@ def operate(operator_config: ForecastOperatorConfig) -> ForecastResults:
             results = None
 
     else:
+        # When AUTO_SELECT_SERIES is specified but target_category_columns is not,
+        # we fall back to AUTO_SELECT behavior.
+        if (
+            operator_config.spec.model == AUTO_SELECT_SERIES
+            and not operator_config.spec.target_category_columns
+        ):
+
+            operator_config.spec.model = AUTO_SELECT
+            model = ForecastOperatorModelFactory.get_model(operator_config, datasets)
         # For other cases, use the single selected model
         results = model.generate_report()
     # saving to model catalog

tests/operators/forecast/test_datasets.py

@@ -12,6 +12,7 @@
 import pandas as pd
 import pytest
 import yaml
+import numpy as np
 
 from ads.opctl.operator.cmd import run
 from ads.opctl.operator.lowcode.forecast.__main__ import operate as forecast_operate
@@ -413,5 +414,44 @@ def run_operator(
 #         generate_train_metrics = True
 
 
+def test_missing_data_autoselect_series():
+    """Test case for auto-select-series with missing data."""
+    data = {
+        "Date": pd.to_datetime(
+            [
+                "2023-01-01",
+                "2023-01-02",
+                "2023-01-03",
+                "2023-01-04",
+                "2023-01-05",
+                "2023-01-06",
+                "2023-01-07",
+                "2023-01-08",
+                "2023-01-09",
+                "2023-01-10",
+            ]
+        ),
+        "Y": [1, 2, np.nan, 4, 5, 6, 7, 8, 9, 10],
+        "Category": ["A", "A", "A", "A", "A", "A", "A", "A", "A", "A"],
+    }
+    df = pd.DataFrame(data)
+
+    with tempfile.TemporaryDirectory() as tmpdirname:
+        output_data_path = f"{tmpdirname}/results"
+        yaml_i = deepcopy(TEMPLATE_YAML)
+        yaml_i["spec"]["model"] = "auto-select-series"
+        yaml_i["spec"]["historical_data"].pop("url")
+        yaml_i["spec"]["historical_data"]["data"] = df
+        yaml_i["spec"]["target_column"] = "Y"
+        yaml_i["spec"]["datetime_column"]["name"] = "Date"
+        yaml_i["spec"]["target_category_columns"] = ["Category"]
+        yaml_i["spec"]["horizon"] = 2
+        yaml_i["spec"]["output_directory"]["url"] = output_data_path
+
+        operator_config = ForecastOperatorConfig.from_dict(yaml_i)
+        forecast_operate(operator_config)
+        check_output_for_errors(output_data_path)
+
+
 if __name__ == "__main__":
     pass