Remove unused levels from categorical index

If `unique_id` is categorical, and some levels were removed, then
`value_counts()` in `_grouped_array_from_df()` still returns these
levels in the output with their counts being 0.

`_grouped_array_from_df()`, however, does not expect that some counts
could be zero, therefore, it makes sense to drop unused levels from a
categorical index.

nbs/core.ipynb

@@ -996,6 +996,11 @@
     "            self._validate_df(df)\n",
     "            if df.index.name != 'unique_id':\n",
     "                df = df.set_index('unique_id')\n",
+    "            if isinstance(df.index, pd.CategoricalIndex):\n",
+    "                # If `unique_id` is categorical and some levels are not present\n",
+    "                # in `df`, then `.count_values()` in `unique_id` does not\n",
+    "                # produce expected output, so drop these unused levels.\n",
+    "                df.index = df.index.remove_unused_categories()\n",
     "            df = _parse_ds_type(df)\n",
     "            self.ga, self.uids, self.last_dates, self.ds = _grouped_array_from_df(df, sort_df)\n",
     "            self.n_jobs = _get_n_jobs(len(self.ga), self.n_jobs)\n",

statsforecast/core.py

@@ -547,6 +547,11 @@ def _prepare_fit(self, df, sort_df):
             self._validate_df(df)
             if df.index.name != "unique_id":
                 df = df.set_index("unique_id")
+            if isinstance(df.index, pd.CategoricalIndex):
+                # If `unique_id` is categorical and some levels are not present
+                # in `df`, then `.count_values()` in `unique_id` does not
+                # produce expected output, so drop these unused levels.
+                df.index = df.index.remove_unused_categories()
             df = _parse_ds_type(df)
             self.ga, self.uids, self.last_dates, self.ds = _grouped_array_from_df(
                 df, sort_df