sklearndf

2.4.2 verified Mon Apr 27 auth: no python

Data frame support and feature traceability for scikit-learn. Version 2.4.2 supports Python >=3.9, <4 and integrates with scikit-learn 1.6+. Release cadence is irregular, with recent minor versions every few months.

pip install sklearndf

Common errors

error ImportError: cannot import name 'EstimatorWrapperDF' from 'sklearndf' ↓

cause EstimatorWrapperDF is not in the top-level sklearndf module.

fix

Use: from sklearndf.transformation import EstimatorWrapperDF

error AttributeError: 'EstimatorWrapperDF' object has no attribute 'feature_names_in_' ↓

cause The estimator has not been fitted or does not expose feature_names_in_ (requires scikit-learn >=1.0).

fix

Ensure the estimator is fitted and scikit-learn version is >=1.0.

error TypeError: All intermediate steps should be transformers or implement fit_transform. 'ClassifierDF' doesn't ↓

cause Using a classifier (ClassifierDF) inside a PipelineDF as an intermediate step instead of a transformer.

fix

Place classifiers only at the end of the pipeline.

error ValueError: could not convert string to float: '...' ↓

cause DataFrame contains non-numeric columns; sklearndf wrappers expect numeric input.

fix

Preprocess categorical features using OneHotEncoder or similar transformer before passing to estimator.

Warnings

breaking sklearndf 2.4.0 introduced scikit-learn 1.7 support. Older scikit-learn versions may break with 2.4.0+. ↓

fix Upgrade scikit-learn to >=1.6 (or 1.7 for full support).

gotcha Wrapper classes only accept scikit-learn estimators that support DataFrame output. Not all sklearn estimators do; check sklearn documentation. ↓

fix Use only estimators that are known to return DataFrames (e.g., from sklearn.ensemble, sklearn.linear_model).

deprecated The pipeline module's `FitDataFrame` and `TransformDataFrame` are deprecated in favor of `EstimatorWrapperDF` and `TransformerWrapperDF`. ↓

fix Use `sklearndf.transformation.EstimatorWrapperDF` for estimators and `TransformerWrapperDF` for transformers.

gotcha When using with sklearn pipelines, use `sklearndf.pipeline.PipelineDF` instead of `sklearn.pipeline.Pipeline` to preserve DataFrame features. ↓

fix Replace `from sklearn.pipeline import Pipeline` with `from sklearndf.pipeline import PipelineDF`.

Install

conda install -c conda-forge -c bcg_gamma sklearndf

Imports

EstimatorWrapperDF

wrong

from sklearndf import EstimatorWrapperDF

correct

from sklearndf.transformation import EstimatorWrapperDF

EstimatorWrapperDF is in sklearndf.transformation, not top-level.

ClassifierDF
wrong
```
from sklearn import ...
```
correct
```
from sklearndf.classification import ClassifierDF
```
Do not import from sklearn directly; use sklearndf wrappers for DataFrame support.

Quickstart

Creates a DecisionTreeClassifier with sklearndf wrapper, fits on iris DataFrame, and prints feature names and predictions.

from sklearn.datasets import load_iris
from sklearn.tree import DecisionTreeClassifier
from sklearndf.classification import ClassifierDF

data = load_iris(as_frame=True)
X, y = data.data, data.target

clf = ClassifierDF(DecisionTreeClassifier())
clf.fit(X, y)
print(clf.feature_names_in_)
print(clf.predict(X[:5]))