scikit-datasets

0.2.5 verified Fri May 01 auth: no python

scikit-datasets provides a collection of ready-to-use datasets compatible with scikit-learn, wrapping common benchmarks like MNIST, CIFAR, and more. Version 0.2.5, updated irregularly.

pip install scikit-datasets

Common errors

error ModuleNotFoundError: No module named 'scikit_datasets' ↓

cause The import package was renamed to 'skdata' in version 0.2.5, but older tutorials use 'scikit_datasets'.

fix

Use 'from skdata import load_...' or downgrade to version 0.1.x (not recommended).

error AttributeError: module 'skdata' has no attribute 'datasets' ↓

cause Users try to access datasets via a submodule that does not exist; functions are directly on the skdata module.

fix

Use 'from skdata import load_mnist' directly, not 'skdata.datasets.load_mnist'.

Warnings

breaking In version 0.2.5, the package name changed from 'scikit_datasets' to 'skdata' for imports. Code using 'from scikit_datasets import ...' will break. ↓

fix Change import to 'from skdata import ...'.

deprecated The 'as_frames' parameter for returning pandas DataFrames is deprecated in favor of 'as_frame=True', which will become the default in future versions. ↓

fix Use 'as_frame=True' instead of 'as_frames=True'.

gotcha Dataset downloads can be large (e.g., MNIST ~15MB, CIFAR-10 ~170MB) and are cached in ~/skdata_data/. Not all datasets are bundled; some require an internet connection on first load. ↓

fix Ensure internet access for first-time download, or cache the data directory.

Imports

load_mnist

wrong

from scikit_datasets import load_mnist

correct

from skdata import load_mnist

Module package is 'skdata' not 'scikit_datasets'.

load_cifar10
wrong
```
import skdata.datasets.load_cifar10
```
correct
```
from skdata import load_cifar10
```
Functions are directly under skdata, not nested.

Quickstart

Loads MNIST and fits an SVM classifier.

from skdata import load_mnist
from sklearn.svm import SVC

data = load_mnist()
X, y = data.data, data.target
clf = SVC().fit(X, y)
print(clf.score(X, y))