Fix export_text single-feature handling (swev-id: scikit-learn__scikit-learn-14053)

[casey-brooks]

Summary

• ensure export_text resolves feature names by index instead of node ids
• avoid TREE_UNDEFINED indices by building names from n_features length
• add regression coverage for single-feature classifiers with and without provided names

Reproduction

from sklearn.tree import DecisionTreeClassifier
from sklearn.tree.export import export_text
from sklearn.datasets import load_iris

X, y = load_iris(return_X_y=True)
X = X[:, 0].reshape(-1, 1)

tree = DecisionTreeClassifier()
tree.fit(X, y)
tree_text = export_text(tree, feature_names=['sepal_length'])
print(tree_text)

Observed failure (pre-fix)

Traceback (most recent call last):
  File "<stdin>", line 18, in <module>
  File "/workspace/tmp_old_export.py", line 893, in export_text
    feature_names_ = [feature_names[i] for i in tree_.feature]
  File "/workspace/tmp_old_export.py", line 893, in <listcomp>
    feature_names_ = [feature_names[i] for i in tree_.feature]
IndexError: list index out of range

Testing

• PYTHONPATH=/workspace/scikit-learn LD_LIBRARY_PATH=/workspace/sklearn-py38/lib:$LD_LIBRARY_PATH /workspace/sklearn-py38/bin/flake8 sklearn/tree/export.py sklearn/tree/tests/test_export.py
• PYTHONPATH=/workspace/sklearn-py38/lib/python3.8/site-packages LD_LIBRARY_PATH=/workspace/sklearn-py38/lib:$LD_LIBRARY_PATH /workspace/sklearn-py38/bin/pytest /workspace/tmp_tests/test_export.py -k export_text

Fixes #34

[noa-lucent]

Fix looks correct. Thanks for adding regression coverage for single-feature trees.