[FIX] Fix three robustness gaps in OpenMLDataset.__repr__, get_data(), and create_dataset

[phantom-712]

Metadata

• Reference Issue: Fixes [FIX] : Three dataset robustness gaps in OpenMLDataset and create_dataset #1711
• New Tests Added: No
• Documentation Updated: Yes (docstring for default_target_attribute in create_dataset)
• Change Log Entry: "Fix KeyError in OpenMLDataset.repr on partial qualities, bare KeyError in get_data() on invalid target, and misleading str annotation on create_dataset default_target_attribute"

Details

What does this PR fix?

Three related robustness gaps in openml/datasets/dataset.py and openml/datasets/functions.py, all verified on openml==0.16.0.

Bug 1 - OpenMLDataset.__repr__ crashes with KeyError on partial or NaN qualities

_get_repr_body_fields accessed _qualities with direct dict indexing. For newly uploaded datasets where the server has only partially computed qualities, repr() crashed with a KeyError when NumberOfFeatures or NumberOfInstances keys were absent or NaN. Fixed by replacing direct indexing with .get() and a pd.isna() guard, missing or NaN quality keys are silently omitted from the repr output.

Bug 2 - get_data() raises a bare pandas KeyError on invalid target

get_data() called data.drop(columns=[target_name]) without first checking whether target_name existed in data.columns. When the column was absent (typo, or silently removed by include_row_id=False / include_ignore_attribute=False), pandas raised a raw KeyError with no OpenML context. Fixed by adding an explicit pre-drop check that raises a ValueError with a clear message listing available columns, and a separate message when the column was filtered out by an OpenML flag.

Bug 3 - create_dataset has a misleading str annotation on default_target_attribute

The REST API allows default_target_attribute to be None for unsupervised datasets. The Python function annotated it as str, causing mypy errors and giving users no indication that None was valid. An existing TODO in the code explicitly acknowledged this mismatch. Fixed by changing the annotation to str | None, updating the docstring to document None as valid for unsupervised use cases, and removing the TODO.

Scope

All changes are contained in two files only:

• openml/datasets/dataset.py - Bugs 1 and 2
• openml/datasets/functions.py - Bug 3

No changes to the REST API surface, no new public methods, no new files.

Reproduction

Bug 1:

import openml
ds = openml.datasets.get_dataset(61, download_data=False, download_qualities=True)
ds._qualities = {"SomeOtherQuality": 1.0}
repr(ds)  # KeyError before fix, clean output after

Bug 2:

import openml
ds = openml.datasets.get_dataset(61, download_data=True)
ds.get_data(target="nonexistent_column")  # bare pandas KeyError before, clear ValueError after

Bug 3:

# mypy flagged create_dataset(..., default_target_attribute=None) as a type error before fix

[codecov-commenter]

Codecov Report

❌ Patch coverage is 68.75000% with 5 lines in your changes missing coverage. Please review.
✅ Project coverage is 53.78%. Comparing base (05a58ee) to head (9cf22c1).

Files with missing lines	Patch %	Lines
openml/datasets/dataset.py	68.75%	5 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1713      +/-   ##
==========================================
+ Coverage   53.75%   53.78%   +0.03%     
==========================================
  Files          63       63              
  Lines        5222     5230       +8     
==========================================
+ Hits         2807     2813       +6     
- Misses       2415     2417       +2     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.