Error with nulls in categorical columns

[SuryaThiru]

Description

Error thrown when loading penguins dataset:
ValueError: Categorical categories cannot be null

Steps/Code to Reproduce

import openml
openml.datasets.get_dataset('penguins')

Expected Results

Dataset loads without issue.

Actual Results

>>> openml.datasets.get_dataset('penguins')
Traceback (most recent call last):
 File "<stdin>", line 1, in <module>
 File "/home/suryak/Projects/sandbox/mlgauge/env/lib/python3.8/site-packages/openml/datasets/functions.py", line 519, in get_dataset
 dataset = _create_dataset_from_description(
 File "/home/suryak/Projects/sandbox/mlgauge/env/lib/python3.8/site-packages/openml/datasets/functions.py", line 1132, in _create_dataset_from_description
 return OpenMLDataset(
 File "/home/suryak/Projects/sandbox/mlgauge/env/lib/python3.8/site-packages/openml/datasets/dataset.py", line 241, in __init__
 ) = self._create_pickle_in_cache(data_file)
 File "/home/suryak/Projects/sandbox/mlgauge/env/lib/python3.8/site-packages/openml/datasets/dataset.py", line 526, in _create_pickle_in_cache
 X, categorical, attribute_names = self._parse_data_from_arff(data_file)
 File "/home/suryak/Projects/sandbox/mlgauge/env/lib/python3.8/site-packages/openml/datasets/dataset.py", line 457, in _parse_data_from_arff
 self._unpack_categories(X[column_name], categories_names[column_name])
 File "/home/suryak/Projects/sandbox/mlgauge/env/lib/python3.8/site-packages/openml/datasets/dataset.py", line 686, in _unpack_categories
 raw_cat = pd.Categorical(col, ordered=True, categories=categories)
 File "/home/suryak/Projects/sandbox/mlgauge/env/lib/python3.8/site-packages/pandas/core/arrays/categorical.py", line 304, in __init__
 dtype = CategoricalDtype._from_values_or_dtype(
 File "/home/suryak/Projects/sandbox/mlgauge/env/lib/python3.8/site-packages/pandas/core/dtypes/dtypes.py", line 273, in _from_values_or_dtype
 dtype = CategoricalDtype(categories, ordered)
 File "/home/suryak/Projects/sandbox/mlgauge/env/lib/python3.8/site-packages/pandas/core/dtypes/dtypes.py", line 160, in __init__
 self._finalize(categories, ordered, fastpath=False)
 File "/home/suryak/Projects/sandbox/mlgauge/env/lib/python3.8/site-packages/pandas/core/dtypes/dtypes.py", line 314, in _finalize
 categories = self.validate_categories(categories, fastpath=fastpath)
 File "/home/suryak/Projects/sandbox/mlgauge/env/lib/python3.8/site-packages/pandas/core/dtypes/dtypes.py", line 508, in validate_categories
 raise ValueError("Categorical categories cannot be null")
ValueError: Categorical categories cannot be null

Versions

Linux-5.9.16-1-MANJARO-x86_64-with-glibc2.10
Python 3.8.5 (default, Sep 4 2020, 07:30:14)
[GCC 7.3.0]
Pandas 1.2.2
NumPy 1.20.1
SciPy 1.6.0
Scikit-Learn 0.24.1
OpenML 0.11.0

[Neeratyoy]

Hi @SuryaThiru

openml.datasets.get_dataset("penguins") appears to work (for me) with the dev branch, with the dependencies (for numpy, pandas especially) as you listed under versions. Could you kindly confirm the same?

If it works, please use the dev branch. We are planning to release a new version soon which then should fix this issue!

[SuryaThiru]

Installed 0.11.1dev using this command on colab (internet a bit slow to clone openml repo locally):

pip install git+https://github.com/openml/openml-python@develop

Now this command seems to work:

import openml
data = openml.datasets.get_dataset('penguins')
data

==============
Name..........: penguins
Version.......: 1
Format........: ARFF
Upload Date...: 2020-07-22 18:34:52
Licence.......: CC0
Download URL..: https://www.openml.org/data/v1/download/21854866/penguins.arff
OpenML URL....: https://www.openml.org/d/42585
# of features.: 7
# of instances: 344

However, this throws the same error

>>> data.get_data()

---------------------------------------------------------------------------
ValueError Traceback (most recent call last)
<ipython-input-10-b4e1f6df3b81> in <module>()
----> 1 data.get_data()

9 frames
/usr/local/lib/python3.7/dist-packages/openml/datasets/dataset.py in get_data(self, target, include_row_id, include_ignore_attribute, dataset_format)
 691 List of attribute names.
 692 """
--> 693 data, categorical, attribute_names = self._load_data()
 694 
 695 to_exclude = []

/usr/local/lib/python3.7/dist-packages/openml/datasets/dataset.py in _load_data(self)
 529 
 530 file_to_load = self.data_file if self.parquet_file is None else self.parquet_file
--> 531 return self._cache_compressed_file_from_file(file_to_load)
 532 
 533 # helper variable to help identify where errors occur

/usr/local/lib/python3.7/dist-packages/openml/datasets/dataset.py in _cache_compressed_file_from_file(self, data_file)
 486 
 487 if data_file.endswith(".arff"):
--> 488 data, categorical, attribute_names = self._parse_data_from_arff(data_file)
 489 elif data_file.endswith(".pq"):
 490 try:

/usr/local/lib/python3.7/dist-packages/openml/datasets/dataset.py in _parse_data_from_arff(self, arff_file_path)
 443 if attribute_dtype[column_name] in ("categorical", "boolean"):
 444 col.append(
--> 445 self._unpack_categories(X[column_name], categories_names[column_name])
 446 )
 447 elif attribute_dtype[column_name] in ("floating", "integer"):

/usr/local/lib/python3.7/dist-packages/openml/datasets/dataset.py in _unpack_categories(series, categories)
 648 # We require two lines to create a series of categories as detailed here:
 649 # https://pandas.pydata.org/pandas-docs/version/0.24/user_guide/categorical.html#series-creation # noqa E501
--> 650 raw_cat = pd.Categorical(col, ordered=True, categories=categories)
 651 return pd.Series(raw_cat, index=series.index, name=series.name)
 652 

/usr/local/lib/python3.7/dist-packages/pandas/core/arrays/categorical.py in __init__(self, values, categories, ordered, dtype, fastpath)
 306 
 307 dtype = CategoricalDtype._from_values_or_dtype(
--> 308 values, categories, ordered, dtype
 309 )
 310 # At this point, dtype is always a CategoricalDtype, but

/usr/local/lib/python3.7/dist-packages/pandas/core/dtypes/dtypes.py in _from_values_or_dtype(cls, values, categories, ordered, dtype)
 275 # Note: This could potentially have categories=None and
 276 # ordered=None.
--> 277 dtype = CategoricalDtype(categories, ordered)
 278 
 279 return dtype

/usr/local/lib/python3.7/dist-packages/pandas/core/dtypes/dtypes.py in __init__(self, categories, ordered)
 162 
 163 def __init__(self, categories=None, ordered: Ordered = False):
--> 164 self._finalize(categories, ordered, fastpath=False)
 165 
 166 @classmethod

/usr/local/lib/python3.7/dist-packages/pandas/core/dtypes/dtypes.py in _finalize(self, categories, ordered, fastpath)
 316 
 317 if categories is not None:
--> 318 categories = self.validate_categories(categories, fastpath=fastpath)
 319 
 320 self._categories = categories

/usr/local/lib/python3.7/dist-packages/pandas/core/dtypes/dtypes.py in validate_categories(categories, fastpath)
 490 
 491 if categories.hasnans:
--> 492 raise ValueError("Categorical categories cannot be null")
 493 
 494 if not categories.is_unique:

ValueError: Categorical categories cannot be null

[PGijsbers]

For those that experience this issue, the dataset is still downloaded to ~\.openml\cache\org\openml\www\datasets\42585. So if you only care about the arff file that might be of help. I'll do my best to set up a PR this afternoon.

[PGijsbers]

Fixed in #1037. Please use the develop branch for now, we plan to do a new PyPI release soon.