fix(calendar): preserve weekly and monthly periods in sparse calendars

qlib/utils/resam.py

@@ -25,6 +25,8 @@ def resam_calendar(
         Frequency of the raw calendar
     freq_sam : str
         Sample frequency
+        Weekly and monthly sampling counts periods present in the input, so entirely
+        missing weeks or months do not advance the sampling count.
     region: str
         Region, for example, "cn", "us"
     Returns
@@ -57,13 +59,13 @@ def resam_calendar(
             return _calendar_day[:: freq_sam.count]
 
         elif freq_sam.base == Freq.NORM_FREQ_WEEK:
-            _day_in_week = np.array(list(map(lambda x: x.dayofweek, _calendar_day)))
-            _calendar_week = _calendar_day[np.ediff1d(_day_in_week, to_begin=-1) < 0]
+            _week = pd.DatetimeIndex(_calendar_day).to_period("W")
+            _calendar_week = _calendar_day[~_week.duplicated()]
             return _calendar_week[:: freq_sam.count]
 
         elif freq_sam.base == Freq.NORM_FREQ_MONTH:
-            _day_in_month = np.array(list(map(lambda x: x.day, _calendar_day)))
-            _calendar_month = _calendar_day[np.ediff1d(_day_in_month, to_begin=-1) < 0]
+            _month = pd.DatetimeIndex(_calendar_day).to_period("M")
+            _calendar_month = _calendar_day[~_month.duplicated()]
             return _calendar_month[:: freq_sam.count]
         else:
             raise ValueError("sampling freq must be xmin, xd, xw, xm")

tests/misc/test_resam.py

@@ -0,0 +1,84 @@
+# Copyright (c) Microsoft Corporation.
+# Licensed under the MIT License.
+
+import numpy as np
+import pandas as pd
+import pytest
+
+from qlib.constant import REG_CN
+from qlib.utils.resam import resam_calendar
+
+
+@pytest.mark.parametrize(
+    "dates, raw_freq, sampled_freq, expected",
+    [
+        (
+            ["2024-01-01", "2024-01-02", "2024-01-09", "2024-01-10", "2024-01-17"],
+            "day",
+            "week",
+            ["2024-01-01", "2024-01-09", "2024-01-17"],
+        ),
+        (
+            ["2023-12-25", "2024-01-01", "2024-01-08", "2024-01-15"],
+            "week",
+            "2week",
+            ["2023-12-25", "2024-01-08"],
+        ),
+        (
+            ["2023-12-04", "2023-12-05", "2024-01-08", "2024-01-09", "2024-02-12"],
+            "day",
+            "month",
+            ["2023-12-04", "2024-01-08", "2024-02-12"],
+        ),
+        (
+            ["2023-12-01", "2024-01-01", "2024-02-01", "2024-03-01"],
+            "month",
+            "2month",
+            ["2023-12-01", "2024-02-01"],
+        ),
+        (
+            ["2023-12-29 09:30", "2023-12-29 13:00", "2024-01-02 09:30", "2024-01-02 13:00"],
+            "1min",
+            "week",
+            ["2023-12-29", "2024-01-02"],
+        ),
+    ],
+)
+def test_resam_calendar_period_boundaries(dates, raw_freq, sampled_freq, expected):
+    calendar = np.array([pd.Timestamp(value) for value in dates])
+    result = resam_calendar(calendar, raw_freq, sampled_freq, region=REG_CN)
+    np.testing.assert_array_equal(result, np.array([pd.Timestamp(value) for value in expected]))
+
+
+@pytest.mark.parametrize(
+    "start, end, gap_start, gap_end, sampled_freq, expected",
+    [
+        (
+            "2024-01-29",
+            "2024-03-08",
+            "2024-02-12",
+            "2024-02-16",
+            "2week",
+            ["2024-01-29", "2024-02-19", "2024-03-04"],
+        ),
+        (
+            "2024-01-01",
+            "2024-06-28",
+            "2024-03-01",
+            "2024-03-31",
+            "2month",
+            ["2024-01-01", "2024-04-01", "2024-06-03"],
+        ),
+    ],
+)
+def test_resam_calendar_counts_nonempty_periods(start, end, gap_start, gap_end, sampled_freq, expected):
+    calendar = pd.bdate_range(start, end)
+    calendar = calendar[(calendar < gap_start) | (calendar > gap_end)]
+    result = resam_calendar(calendar.to_pydatetime(), "day", sampled_freq, region=REG_CN)
+    np.testing.assert_array_equal(result, np.array([pd.Timestamp(value) for value in expected]))
+
+
+@pytest.mark.parametrize("sampled_freq", ["week", "month", "2week", "2month"])
+def test_resam_calendar_empty(sampled_freq):
+    calendar = np.array([], dtype=object)
+    np.testing.assert_array_equal(resam_calendar(calendar, "day", sampled_freq, region=REG_CN), calendar)