From 7815dd7fc69171b89d39a49bf5308efb8fe7876a Mon Sep 17 00:00:00 2001
From: Oleksandr Shchur <shchuro@amazon.com>
Date: Tue, 15 Nov 2022 19:57:06 +0100
Subject: [PATCH] Fix leaderboard with static features (#2398)

---
 .../src/autogluon/timeseries/learner.py       |  2 +-
 timeseries/tests/unittests/test_learner.py    | 34 +++++++++++++++++++
 2 files changed, 35 insertions(+), 1 deletion(-)

diff --git a/timeseries/src/autogluon/timeseries/learner.py b/timeseries/src/autogluon/timeseries/learner.py
index 197f7929c27..3ff49079663 100644
--- a/timeseries/src/autogluon/timeseries/learner.py
+++ b/timeseries/src/autogluon/timeseries/learner.py
@@ -375,7 +375,7 @@ def score(
         return self.load_trainer().score(data=data, model=model, metric=metric)
 
     def leaderboard(self, data: Optional[TimeSeriesDataFrame] = None) -> pd.DataFrame:
-        if self.static_feature_pipeline.is_fit():
+        if data is not None and self.static_feature_pipeline.is_fit():
             fix_message = (
                 "Please make sure that data has static_features with columns and dtypes exactly matching "
                 "train_data.static_features. "
diff --git a/timeseries/tests/unittests/test_learner.py b/timeseries/tests/unittests/test_learner.py
index 33fe013f520..8e704466a1f 100644
--- a/timeseries/tests/unittests/test_learner.py
+++ b/timeseries/tests/unittests/test_learner.py
@@ -394,3 +394,37 @@ def test_when_ignore_index_is_true_and_known_covariates_available_then_learner_c
     known_covariates = get_data_frame_with_variable_lengths(ITEM_ID_TO_LENGTH, known_covariates_names=["X", "Y"])
     preds = learner.predict(train_data, known_covariates=known_covariates)
     assert preds.item_ids.equals(train_data.item_ids)
+
+
+@pytest.mark.parametrize("pred_data_present", [True, False])
+@pytest.mark.parametrize("static_features_present", [True, False])
+@pytest.mark.parametrize("known_covariates_present", [True, False])
+def test_when_train_data_has_static_or_dynamic_feat_then_leaderboard_works(
+    temp_model_path, pred_data_present, static_features_present, known_covariates_present
+):
+    if static_features_present:
+        static_features = get_static_features(["B", "A"], feature_names=["f1", "f2"])
+    else:
+        static_features = None
+
+    if known_covariates_present:
+        known_covariates_names = ["X", "Y"]
+    else:
+        known_covariates_names = None
+
+    train_data = get_data_frame_with_variable_lengths(
+        {"B": 20, "A": 15}, static_features=static_features, known_covariates_names=known_covariates_names
+    )
+
+    if pred_data_present:
+        pred_data = get_data_frame_with_variable_lengths(
+            {"B": 20, "A": 15}, static_features=static_features, known_covariates_names=known_covariates_names
+        )
+    else:
+        pred_data = None
+
+    learner = TimeSeriesLearner(path_context=temp_model_path)
+    learner.fit(train_data=train_data, hyperparameters=HYPERPARAMETERS_DUMMY)
+    leaderboard = learner.leaderboard(data=pred_data)
+    assert len(leaderboard) > 0
+    assert ("score_test" in leaderboard.columns) == pred_data_present