feat: enhance convert_index_to_timestamps to infer and set frequency for DatetimeIndex

Author: dwr-psandhu

pydsm/analysis/postpro.py

@@ -35,6 +35,10 @@ def convert_index_to_timestamps(df):
     if isinstance(df.index, pd.core.indexes.datetimes.DatetimeIndex):
         if df.index.dtype != "datetime64[ns]":
             df.index = df.index.astype("datetime64[ns]")
+            if df.index.freq is None:
+                inferred = pd.infer_freq(df.index)
+                if inferred is not None:
+                    df.index.freq = pd.tseries.frequencies.to_offset(inferred)
         return
     df.index = df.index.astype("datetime64[ns]")
     df.index.freq = pd.infer_freq(df.index)  # if possible

tests/test_postpro.py

@@ -2,6 +2,7 @@
 import pytest
 import numpy as np
 import pandas as pd
+from pydsm.analysis.postpro import convert_index_to_timestamps, merge
 
 
 def test_postprocessor():
@@ -23,3 +24,72 @@ def test_postprocessor():
     assert p2.df.index[0] == pd.to_datetime("03OCT2016 0000")
     assert p2.df.index[-1] == pd.to_datetime("05OCT2016 0000")
     # assert_frame_equal(p.df, p2.df,check_names=False, check_column_type=False, check_like=False)
+
+
+# ---------------------------------------------------------------------------
+# convert_index_to_timestamps — datetime resolution normalisation
+# ---------------------------------------------------------------------------
+
+class TestConvertIndexToTimestamps:
+    """Regression: convert_index_to_timestamps must coerce datetime64[us] to
+    datetime64[ns] — previously it returned early without coercing because the
+    index was already a DatetimeIndex."""
+
+    def _make(self, dtype, n=10):
+        idx = pd.date_range("2020-01-01", periods=n, freq="15min").astype(dtype)
+        return pd.DataFrame({"v": range(n)}, index=idx)
+
+    def test_ns_index_unchanged(self):
+        df = self._make("datetime64[ns]")
+        convert_index_to_timestamps(df)
+        assert df.index.dtype == np.dtype("datetime64[ns]")
+
+    def test_us_index_coerced_to_ns(self):
+        """Before the fix this was a no-op; the index stayed as datetime64[us]."""
+        df = self._make("datetime64[us]")
+        convert_index_to_timestamps(df)
+        assert df.index.dtype == np.dtype("datetime64[ns]"), (
+            f"Expected datetime64[ns], got {df.index.dtype}"
+        )
+
+    def test_period_index_converted(self):
+        idx = pd.period_range("2020-01-01", periods=10, freq="15min")
+        df = pd.DataFrame({"v": range(10)}, index=idx)
+        convert_index_to_timestamps(df)
+        assert isinstance(df.index, pd.DatetimeIndex)
+
+
+# ---------------------------------------------------------------------------
+# merge() — mixed datetime resolution (ns vs us)
+# ---------------------------------------------------------------------------
+
+class TestMergeMixedResolution:
+    """merge() calls combine_first() which requires aligned indices.
+    Both inputs must be normalised to the same resolution first."""
+
+    def _make(self, dtype, start="2020-01-01", n=20, col="v"):
+        idx = pd.date_range(start, periods=n, freq="15min").astype(dtype)
+        return pd.DataFrame({col: np.random.default_rng(0).random(n)}, index=idx)
+
+    def test_ns_plus_us_does_not_raise(self):
+        df_ns = self._make("datetime64[ns]", start="2020-01-01")
+        df_us = self._make("datetime64[us]", start="2020-01-01 05:00")  # overlapping
+        result = merge([df_ns, df_us])
+        assert result is not None
+        assert isinstance(result.index, pd.DatetimeIndex)
+        assert result.index.dtype == np.dtype("datetime64[ns]")
+
+    def test_both_us_does_not_raise(self):
+        df1 = self._make("datetime64[us]", start="2020-01-01")
+        df2 = self._make("datetime64[us]", start="2020-01-01 05:00")
+        result = merge([df1, df2])
+        assert result is not None
+        assert result.index.dtype == np.dtype("datetime64[ns]")
+
+    def test_result_spans_both_inputs(self):
+        df_ns = self._make("datetime64[ns]", start="2020-01-01", n=4)
+        df_us = self._make("datetime64[us]", start="2020-01-01 01:00", n=4)
+        result = merge([df_ns, df_us])
+        # Result should cover the full range of both inputs
+        assert result.index.min() <= pd.Timestamp("2020-01-01")
+        assert result.index.max() >= pd.Timestamp("2020-01-01 01:45")