fix: reject nulls in id and time columns in validate_format

tests/test_processing.py

@@ -1243,6 +1243,21 @@ def test_none_target():
         to_numpy(series_pd.drop(columns=["unique_id", "ds"])),
     )
 
+
+@pytest.mark.parametrize("engine", ["pandas", "polars"])
+def test_process_df_null_ids(engine):
+    constructor = pd.DataFrame if engine == "pandas" else pl.DataFrame
+    df = constructor(
+        {
+            "unique_id": ["a", "a", None, "b"],
+            "ds": [1, 2, 1, 1],
+            "y": [1.0, 2.0, 3.0, 4.0],
+        }
+    )
+    with pytest.raises(ValueError, match="id column .* contains null values"):
+        process_df(df, "unique_id", "ds", "y")
+
+
 @pytest.mark.parametrize("n_static_features", [0, 2])
 def test_n_static_features(static_features, n_static_features):
     series_pl = generate_series(

tests/test_validation.py

@@ -1,7 +1,9 @@
 import datetime
 
+import numpy as np
 import pandas as pd
 import polars as pl
+import pytest
 from conftest import assert_raises_with_message
 
 from utilsforecast.compat import POLARS_INSTALLED
@@ -71,6 +73,48 @@ def test_validate_format():
             assert_raises_with_message(validate_format, "('sales') should have a numeric data type", df, target_col='sales')
 
 
+@pytest.mark.parametrize(
+    "ids",
+    [
+        pd.Series(["a", None]),
+        pd.Series(["a", np.nan]),
+        pd.Series(["a", None], dtype="string[pyarrow]"),
+        pd.Series(["a", np.nan], dtype="category"),
+        pd.Series([1, None], dtype="Int64"),
+        pd.Series([1.0, np.nan]),
+        pl.Series(["a", None]),
+        pl.Series(["a", None], dtype=pl.Categorical),
+        pl.Series([1, None]),
+    ],
+)
+@pytest.mark.parametrize("target_col", ["y", None])
+def test_validate_format_null_ids(ids, target_col):
+    constructor = pd.DataFrame if isinstance(ids, pd.Series) else pl.DataFrame
+    df = constructor({"uid": ids, "ds": [1, 2], "y": [1.0, 2.0]})
+    with pytest.raises(ValueError, match=r"The id column \('uid'\) contains null"):
+        validate_format(df, id_col="uid", target_col=target_col)
+
+
+@pytest.mark.parametrize(
+    "times",
+    [
+        pd.Series(pd.to_datetime(["2000-01-01", None])),
+        pd.Series([datetime.date(2000, 1, 1), None], dtype="date32[pyarrow]"),
+        pd.Series([1, None], dtype="Int64"),
+        pd.Series([1, None]),
+        pl.Series([datetime.datetime(2000, 1, 1), None]),
+        pl.Series([datetime.date(2000, 1, 1), None]),
+        pl.Series([1, None]),
+    ],
+)
+@pytest.mark.parametrize("target_col", ["y", None])
+def test_validate_format_null_times(times, target_col):
+    constructor = pd.DataFrame if isinstance(times, pd.Series) else pl.DataFrame
+    df = constructor({"unique_id": [1, 1], "time": times, "y": [1.0, 2.0]})
+    with pytest.raises(ValueError, match=r"The time column \('time'\) contains null"):
+        validate_format(df, time_col="time", target_col=target_col)
+
+
 def test_validate_freq():
     assert_raises_with_message(validate_freq, 'provide a valid integer',pd.Series([1, 2]), 'D')
     assert_raises_with_message(validate_freq, 'provide a valid pandas or polars offset', pd.to_datetime(['2000-01-01']).to_series(), 1),

utilsforecast/validation.py

@@ -39,6 +39,12 @@ def _is_dt_or_int(s: Series) -> bool:
     return _is_dt_dtype(s) or _is_int_dtype(s)
 
 
+def _has_nulls(s: Series) -> bool:
+    if isinstance(s, pd.Series):
+        return s.hasnans
+    return s.null_count() > 0
+
+
 def ensure_shallow_copy(df: pd.DataFrame) -> pd.DataFrame:
     from packaging.version import Version
 
@@ -106,6 +112,12 @@ def validate_format(
     if missing_cols:
         raise ValueError(f"The following columns are missing: {missing_cols}")
 
+    # nulls
+    if _has_nulls(df[id_col]):
+        raise ValueError(f"The id column ('{id_col}') contains null values.")
+    if _has_nulls(df[time_col]):
+        raise ValueError(f"The time column ('{time_col}') contains null values.")
+
     # time col
     if not _is_dt_or_int(df[time_col]):
         times_dtype = df[time_col].dtype