From 8e9d419dcb283a9bc896d32089919834d9dc8e13 Mon Sep 17 00:00:00 2001
From: Martin Durant <martindurant@users.noreply.github.com>
Date: Tue, 10 Oct 2023 09:49:18 -0400
Subject: [PATCH] Allow categorical column with no categories (#888)

---
 fastparquet/core.py                | 7 ++++++-
 fastparquet/dataframe.py           | 2 +-
 fastparquet/test/test_dataframe.py | 9 ++++++++-
 3 files changed, 15 insertions(+), 3 deletions(-)

diff --git a/fastparquet/core.py b/fastparquet/core.py
index a2156c6f..10e70c00 100644
--- a/fastparquet/core.py
+++ b/fastparquet/core.py
@@ -446,7 +446,12 @@ def read_col(column, schema_helper, infile, use_cat=False,
         column.
     """
     cmd = column.meta_data
-    se = schema_helper.schema_element(cmd.path_in_schema)
+    try:
+        se = schema_helper.schema_element(cmd.path_in_schema)
+    except KeyError:
+        # column not present in this row group
+        assign[:] = None
+        return
     off = min((cmd.dictionary_page_offset or cmd.data_page_offset,
                cmd.data_page_offset))
 
diff --git a/fastparquet/dataframe.py b/fastparquet/dataframe.py
index b9341a15..690ea0ca 100644
--- a/fastparquet/dataframe.py
+++ b/fastparquet/dataframe.py
@@ -191,7 +191,7 @@ def set_cats(values, i=i, col=col, **kwargs):
         shape[-1] = size
 
         if isinstance(bvalues, Categorical):
-            code = np.zeros(shape=shape, dtype=bvalues.codes.dtype)
+            code = np.full(fill_value=-1, shape=shape, dtype=bvalues.codes.dtype)
 
             values = Categorical.from_codes(codes=code, dtype=bvalues.dtype)
 
diff --git a/fastparquet/test/test_dataframe.py b/fastparquet/test/test_dataframe.py
index 5f586182..24da294c 100644
--- a/fastparquet/test/test_dataframe.py
+++ b/fastparquet/test/test_dataframe.py
@@ -1,6 +1,7 @@
 import warnings
 from unittest import mock
 
+import numpy as np
 import pandas as pd
 import pytest
 from numpy import empty as np_empty
@@ -26,7 +27,7 @@ def test_empty():
     df, views = empty('category', size=n, cols=['c'],
                       cats={'c': ['one', 'two']})
     views['c'][0] = 1
-    assert df.c[:2].tolist() == ['two', 'one']
+    assert df.c[:2].tolist() == ['two', np.nan]
 
     df, views = empty('i4,i8,f8,f8,O', size=n,
                       cols=['i4', 'i8', 'f8_1', 'f8_2', 'O'])
@@ -34,6 +35,12 @@ def test_empty():
     assert len(views) == 5
 
 
+def test_no_cats():
+    df, views = empty('category', size=10, cols=['c'],
+                      cats={'c': []})
+    assert (views["c"] == -1).all()
+
+
 def test_empty_tz_utc():
     with warnings.catch_warnings():
         warnings.simplefilter("error")