support hours in fill_gaps (#73)

Nixtla · Mar 26, 2024 · c4d6229 · c4d6229
1 parent a094f50
commit c4d6229
Show file tree

Hide file tree

Showing 4 changed files with 20 additions and 16 deletions.
diff --git a/nbs/preprocessing.ipynb b/nbs/preprocessing.ipynb
@@ -178,14 +178,16 @@
     "        return grid.join(df, on=[id_col, time_col], how='left')\n",
     "    if isinstance(freq, str):\n",
     "        offset = pd.tseries.frequencies.to_offset(freq)\n",
-    "        if 'min' in freq:\n",
+    "        if isinstance(offset.base, pd.offsets.Minute):\n",
     "            # minutes are represented as 'm' in numpy\n",
-    "            freq = freq.replace('min', 'm')\n",
-    "        elif 'B' in freq:\n",
-    "            # business day\n",
-    "            if freq != 'B':\n",
+    "            freq = 'm'\n",
+    "        elif isinstance(offset.base, pd.offsets.BusinessDay):\n",
+    "            if offset.n != 1:\n",
     "                raise NotImplementedError('Multiple of a business day')\n",
     "            freq = 'D'\n",
+    "        elif isinstance(offset.base, pd.offsets.Hour):\n",
+    "            # hours are represented as 'h' in numpy\n",
+    "            freq = 'h'\n",
     "        if offset.n > 1:\n",
     "            freq = freq.replace(str(offset.n), '')\n",
     "        try:\n",
@@ -208,7 +210,7 @@
     "    )\n",
     "    uids = np.repeat(times_by_id.index, sizes)    \n",
     "    if isinstance(freq, str):\n",
-    "        if offset.base.name == 'B':\n",
+    "        if isinstance(offset.base, pd.offsets.BusinessDay):\n",
     "            # data was generated daily, we need to keep only business days \n",
     "            bdays = np.is_busday(times)\n",
     "            uids = uids[bdays]\n",
@@ -1620,11 +1622,11 @@
     "        assert max_dates[0] == expected_end\n",
     "\n",
     "n_periods = 100\n",
-    "freqs = ['YE', 'YS', 'ME', 'MS', 'W', 'W-TUE', 'D', 's', 'ms', 1, 2, '20D', '30s', '2YE', '3YS', '30min', 'B']\n",
+    "freqs = ['YE', 'YS', 'ME', 'MS', 'W', 'W-TUE', 'D', 's', 'ms', 1, 2, '20D', '30s', '2YE', '3YS', '30min', 'B', '1h']\n",
     "try:\n",
     "    pd.tseries.frequencies.to_offset('YE')\n",
     "except ValueError:\n",
-    "    freqs = [f.replace('YE', 'Y').replace('ME', 'M') for f in freqs if isinstance(f, str)]\n",
+    "    freqs = [f.replace('YE', 'Y').replace('ME', 'M').replace('h', 'H') for f in freqs if isinstance(f, str)]\n",
     "for freq in freqs:\n",
     "    if isinstance(freq, (pd.offsets.BaseOffset, str)):         \n",
     "        dates = pd.date_range('1900-01-01', periods=n_periods, freq=freq)\n",

diff --git a/settings.ini b/settings.ini
@@ -1,7 +1,7 @@
 [DEFAULT]
 repo = utilsforecast
 lib_name = utilsforecast
-version = 0.1.1
+version = 0.1.2
 min_python = 3.8
 license = apache2
 black_formatting = True

diff --git a/utilsforecast/__init__.py b/utilsforecast/__init__.py
@@ -1 +1 @@
-__version__ = "0.1.1"
+__version__ = "0.1.2"
diff --git a/utilsforecast/preprocessing.py b/utilsforecast/preprocessing.py
@@ -124,14 +124,16 @@ def fill_gaps(
         return grid.join(df, on=[id_col, time_col], how="left")
     if isinstance(freq, str):
         offset = pd.tseries.frequencies.to_offset(freq)
-        if "min" in freq:
+        if isinstance(offset.base, pd.offsets.Minute):
             # minutes are represented as 'm' in numpy
-            freq = freq.replace("min", "m")
-        elif "B" in freq:
-            # business day
-            if freq != "B":
+            freq = "m"
+        elif isinstance(offset.base, pd.offsets.BusinessDay):
+            if offset.n != 1:
                 raise NotImplementedError("Multiple of a business day")
             freq = "D"
+        elif isinstance(offset.base, pd.offsets.Hour):
+            # hours are represented as 'h' in numpy
+            freq = "h"
         if offset.n > 1:
             freq = freq.replace(str(offset.n), "")
         try:
@@ -152,7 +154,7 @@ def fill_gaps(
     )
     uids = np.repeat(times_by_id.index, sizes)
     if isinstance(freq, str):
-        if offset.base.name == "B":
+        if isinstance(offset.base, pd.offsets.BusinessDay):
             # data was generated daily, we need to keep only business days
             bdays = np.is_busday(times)
             uids = uids[bdays]