perf(core): vectorize the date conversion in basemodel_to_df (#7622)

* perf(core): vectorize the date conversion in basemodel_to_df

basemodel_to_df ran three per-element Python passes over the date column of
every payload it converted. Series.apply(to_datetime) calls pandas.to_datetime
once per row, and because each call receives a lone scalar with no format hint,
pandas also re-runs its datetime format guesser on every row.

Parse the column in one call instead, and use vectorized comparisons for the
all-midnight test and the reduction to datetime.date.

The element-wise parse is kept as a fallback for one case that the vectorized
path cannot represent: a column mixing UTC offsets, such as a series crossing a
daylight-savings boundary. Parsing per element lets each value keep its own
offset and yields an object column of tz-aware Timestamps, whereas a single
to_datetime call over that column raises "Mixed timezones detected". That
behaviour is covered by test_to_df_daylight_savings and is preserved exactly.

Measured on a payload built through EquityHistoricalData, median of 9 under an
exclusive CPU lock:

  250 daily bars     8.92 ms -> 1.46 ms
  1,000 daily bars  32.93 ms -> 3.13 ms
  5,000 daily bars 162.25 ms -> 12.23 ms

* fix a broken test - unrelated to this PR...

---------

Co-authored-by: Danglewood <85772166+deeleeramone@users.noreply.github.com>
Co-authored-by: deeleeramone <>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

openbb_platform/core/openbb_core/app/utils/__init__.py

@@ -46,9 +46,22 @@ def basemodel_to_df(
 
     # If the date column contains dates only, convert them to a date to avoid encoding time data.
     if "date" in df.columns:
-        df["date"] = df["date"].apply(to_datetime)
-        if all(t.time() == time(0, 0) for t in df["date"]):
-            df["date"] = df["date"].apply(lambda x: x.date())
+        try:
+            # Parse the column in a single call instead of once per row.
+            parsed = to_datetime(df["date"])
+        except ValueError:
+            # A column that mixes UTC offsets - a series crossing a daylight
+            # savings boundary, for instance - has no single datetime64
+            # representation, so fall back to parsing each value on its own.
+            parsed = df["date"].apply(to_datetime)
+
+        df["date"] = parsed
+
+        if parsed.dtype == object:
+            if all(t.time() == time(0, 0) for t in parsed):
+                df["date"] = parsed.apply(lambda x: x.date())
+        elif (parsed == parsed.dt.normalize()).all():
+            df["date"] = parsed.dt.date
 
     if index and index in df.columns:
         if index == "date":

openbb_platform/core/tests/app/test_utils.py

@@ -196,6 +196,54 @@ def test_basemodel_to_df_with_date_index():
     assert df.index.name == "date"
 
 
+def test_basemodel_to_df_mixed_offsets_all_midnight_reduces_to_date():
+    """A mixed-offset column that is entirely midnight still reduces to dates.
+
+    The payload crosses a daylight-savings boundary, so it has no single
+    datetime64 representation and stays object dtype. Every value is midnight
+    local time, so it must still be reduced to ``datetime.date``.
+    """
+    from datetime import date
+
+    data = [
+        Data(date="2023-11-03 00:00:00-04:00", value=1),
+        Data(date="2023-11-06 00:00:00-05:00", value=2),
+        Data(date="2023-11-08 00:00:00-05:00", value=3),
+    ]
+    result = basemodel_to_df(data)
+    assert list(result["date"]) == [
+        date(2023, 11, 3),
+        date(2023, 11, 6),
+        date(2023, 11, 8),
+    ]
+
+
+def test_basemodel_to_df_falls_back_to_per_element_parse(monkeypatch):
+    """A column the vectorized parse rejects is parsed one value at a time.
+
+    Whether a mixed-offset column raises depends on the pandas version - 3.x
+    raises, 2.x returns object dtype - so the fallback is forced here rather
+    than left to the installed version.
+    """
+    from datetime import date
+
+    real_to_datetime = pd.to_datetime
+
+    def fake_to_datetime(arg, *args, **kwargs):
+        if isinstance(arg, pd.Series):
+            raise ValueError("Mixed timezones detected. Pass utc=True")
+        return real_to_datetime(arg, *args, **kwargs)
+
+    monkeypatch.setattr(pd, "to_datetime", fake_to_datetime)
+
+    data = [
+        Data(date="2023-11-03 00:00:00-04:00", value=1),
+        Data(date="2023-11-06 00:00:00-05:00", value=2),
+    ]
+    result = basemodel_to_df(data)
+    assert list(result["date"]) == [date(2023, 11, 3), date(2023, 11, 6)]
+
+
 def test_df_to_basemodel_with_date_only_column_keeps_date_format():
     df = pd.DataFrame(
         {