fix: retain model names containing interval-like text in evaluation

tests/test_evaluation_model_names.py

@@ -0,0 +1,42 @@
+import pandas as pd
+import polars as pl
+import pytest
+
+from utilsforecast.evaluation import evaluate
+from utilsforecast.losses import coverage, mae
+
+
+@pytest.mark.parametrize("engine", ["pandas", "polars"])
+@pytest.mark.parametrize(
+    "model", ["model-hi-2025-v2", "baseline-lo-10-tuned", "plain-model"]
+)
+@pytest.mark.parametrize("level", [None, 90, 90.5, 1e-5])
+def test_evaluate_infers_models_with_interval_like_names(engine, model, level):
+    data = {
+        "unique_id": ["a", "a"],
+        "ds": [1, 2],
+        "y": [10.0, 20.0],
+        "baseline": [10.0, 18.0],
+        model: [12.0, 18.0],
+    }
+    models = ["baseline", model]
+    metrics = [mae]
+    if level is not None:
+        for name in models:
+            data[f"{name}-lo-{level}"] = [8.0, 17.0]
+            data[f"{name}-hi-{level}"] = [14.0, 22.0]
+        metrics.append(coverage)
+        level = [level]
+    df = pd.DataFrame(data) if engine == "pandas" else pl.DataFrame(data)
+
+    inferred = evaluate(df, metrics=metrics, level=level)
+    explicit = evaluate(df, metrics=metrics, models=models, level=level)
+
+    assert set(inferred.columns) == {"unique_id", "metric", *models}
+    if engine == "polars":
+        inferred = inferred.to_pandas()
+        explicit = explicit.to_pandas()
+    pd.testing.assert_frame_equal(inferred, explicit)
+    errors = inferred.loc[inferred["metric"] == "mae"]
+    assert errors[model].iloc[0] == 2.0
+    assert errors["baseline"].iloc[0] == 1.0

utilsforecast/evaluation.py

@@ -61,7 +61,7 @@ def _get_model_cols(
         c
         for c in cols
         if c not in [id_col, time_col, target_col, cutoff_col]
-        and not re.search(r"-(?:lo|hi)-\d+", c)
+        and not re.search(r"-(?:lo|hi)-\d+(?:\.\d+)?(?:[eE][+-]?\d+)?$", c)
     ]