imbalanced-learn

0.14.1 · active · verified Sun Apr 12

imbalanced-learn is a Python library that provides a comprehensive suite of resampling techniques to address imbalanced datasets in machine learning, where one class significantly outnumbers another. It offers methods for over-sampling (e.g., SMOTE, ADASYN), under-sampling (e.g., NearMiss, EditedNearestNeighbours), and combined approaches, along with ensemble methods tailored for imbalanced data. It is compatible with scikit-learn and is part of scikit-learn-contrib projects. The current stable version is 0.14.1, with regular maintenance releases to ensure compatibility with scikit-learn and Python versions.

Warnings

breaking The `ratio` and `return_indices` parameters have been removed from all samplers. Users should now use the `sampling_strategy` parameter instead. This can break code written for older versions.
Fix: Replace `ratio` and `return_indices` with `sampling_strategy` (e.g., `sampling_strategy='auto'`, `'minority'`, `'not majority'`, or a dictionary specifying target counts).
breaking The `imblearn.ensemble.BalanceCascade` and `imblearn.ensemble.EasyEnsemble` classes were removed after two deprecation cycles. Code relying on these will fail.
Fix: Migrate to alternative ensemble methods or updated API for similar functionality.
deprecated The `n_jobs` parameter in `imblearn.under_sampling.ClusterCentroids` is deprecated due to its deprecation in `sklearn.cluster.KMeans` (which it relies on).
Fix: Avoid using the `n_jobs` parameter in `ClusterCentroids` and rely on scikit-learn's global joblib configuration or alternative parallelization strategies if needed.
gotcha The `imblearn.pipeline.Pipeline`'s internal `check_is_fitted` mechanism for checking if the pipeline is fitted will change from a warning to an error in version 0.15. This change aligns with scikit-learn's behavior.
Fix: Ensure your pipeline is properly fitted before calling methods like `transform` or `predict` to avoid future errors. This often means calling `.fit()` before any other operations on the pipeline.
gotcha imbalanced-learn has strong compatibility requirements with specific `scikit-learn` versions. Installing incompatible versions can lead to `FutureWarning`s or runtime errors.
Fix: Always check the official documentation for the precise `scikit-learn` version compatibility for your `imbalanced-learn` version. Upgrade both packages concurrently if issues arise.

Install

pip install imbalanced-learn PyPI
conda install -c conda-forge imbalanced-learn Conda

Imports

imblearn
```
import imblearn
```

SMOTE

from imblearn.over_sampling import SMOTE

RandomOverSampler

from imblearn.over_sampling import RandomOverSampler

RandomUnderSampler

from imblearn.under_sampling import RandomUnderSampler

Pipeline
```
from imblearn.pipeline import Pipeline
```
imblearn
A common mistake is trying to 'pip install imblearn' directly, which is an old, empty PyPI package. The correct package name for installation is 'imbalanced-learn', though it's imported as 'imblearn'.

Quickstart

This quickstart demonstrates how to use SMOTE (Synthetic Minority Over-sampling Technique) to balance an imbalanced dataset generated with scikit-learn. It first creates a dataset with a 90:10 class distribution and then applies SMOTE to equalize the number of samples in both classes.

from sklearn.datasets import make_classification
from collections import Counter
from imblearn.over_sampling import SMOTE

# Generate an imbalanced dataset
X, y = make_classification(
    n_samples=1000, n_features=2, n_informative=2, n_redundant=0,
    n_repeated=0, n_classes=2, n_clusters_per_class=1,
    weights=[0.9, 0.1], flip_y=0, random_state=42
)

print(f"Original dataset shape: {Counter(y)}")

# Apply SMOTE to oversample the minority class
smote = SMOTE(random_state=42)
X_resampled, y_resampled = smote.fit_resample(X, y)

print(f"Resampled dataset shape: {Counter(y_resampled)}")

view raw JSON →