BUG: Fix float32 precision issues in pd.to_datetime

doc/source/whatsnew/v3.0.0.rst

@@ -626,6 +626,7 @@ Datetimelike
 - Bug in :meth:`DatetimeIndex.union` and :meth:`DatetimeIndex.intersection` when ``unit`` was non-nanosecond (:issue:`59036`)
 - Bug in :meth:`Series.dt.microsecond` producing incorrect results for pyarrow backed :class:`Series`. (:issue:`59154`)
 - Bug in :meth:`to_datetime` not respecting dayfirst if an uncommon date string was passed. (:issue:`58859`)
+- Bug in :meth:`to_datetime` on float32 df with year, month, day etc. columns leads to precision issues and incorrect result. (:issue:`60506`)
 - Bug in :meth:`to_datetime` reports incorrect index in case of any failure scenario. (:issue:`58298`)
 - Bug in :meth:`to_datetime` wrongly converts when ``arg`` is a ``np.datetime64`` object with unit of ``ps``. (:issue:`60341`)
 - Bug in setting scalar values with mismatched resolution into arrays with non-nanosecond ``datetime64``, ``timedelta64`` or :class:`DatetimeTZDtype` incorrectly truncating those scalars (:issue:`56410`)

pandas/core/tools/datetimes.py

@@ -44,6 +44,7 @@
 from pandas.core.dtypes.common import (
     ensure_object,
     is_float,
+    is_float_dtype,
     is_integer,
     is_integer_dtype,
     is_list_like,
@@ -1153,6 +1154,10 @@ def coerce(values):
         # we allow coercion to if errors allows
         values = to_numeric(values, errors=errors)
 
+        # prevent prevision issues in case of float32 # GH#60506
+        if is_float_dtype(values.dtype):
+            values = values.astype("float64")
+
         # prevent overflow in case of int8 or int16
         if is_integer_dtype(values.dtype):
             values = values.astype("int64")

pandas/tests/tools/test_to_datetime.py

@@ -2084,6 +2084,18 @@ def test_dataframe_str_dtype(self, df, cache):
         )
         tm.assert_series_equal(result, expected)
 
+    def test_dataframe_float32_dtype(self, df, cache):
+        # GH#60506
+        # coerce to float64
+        result = to_datetime(df.astype(np.float32), cache=cache)
+        expected = Series(
+            [
+                Timestamp("20150204 06:58:10.001002003"),
+                Timestamp("20160305 07:59:11.001002003"),
+            ]
+        )
+        tm.assert_series_equal(result, expected)
+
     def test_dataframe_coerce(self, cache):
         # passing coerce
         df2 = DataFrame({"year": [2015, 2016], "month": [2, 20], "day": [4, 5]})