Skip to content

Latest commit

ย 

History

History
502 lines (425 loc) ยท 37.6 KB

README.md

File metadata and controls

502 lines (425 loc) ยท 37.6 KB
Raw

IMBENS: Class-imbalanced Ensemble Learning in Python

Status CircleCI Status Read the Docs
PyPI
Traffic
Documentation
Paper & Citation
Language

โณQuick Start with our 5-minute Guide & Detailed Examples

IMBENS (imported as imbens) is a Python library for quick implementation, modification, evaluation, and visualization of ensemble learning from class-imbalanced data. Currently, IMBENS includes over 15 ensemble imbalanced learning algorithms (SMOTEBoost, SMOTEBagging, RUSBoost, EasyEnsemble, SelfPacedEnsemble, etc) and 19 over-/under-sampling methods (SMOTE, ADASYN, TomekLinks, etc) from imbalance-learn.

๐ŸŒˆ IMBENS Highlights

  • ๐Ÿง‘โ€๐Ÿ’ป Ease-of-use: Unified, easy-to-use APIs with documentation and examples.
  • ๐Ÿš€ Performance: Optimized performance with parallelization using joblib.
  • ๐Ÿ“Š Benchmarking: Running & comparing multiple models with our visualizer.
  • ๐Ÿ“บ Monitoring: Powerful, customizable, interactive training logging.
  • ๐Ÿช Versatility: Full compatibility with scikit-learn and imbalanced-learn.
  • ๐Ÿ“ˆ Functionality: Extending existing techniques from binary to multi-class setting.

โœ‚๏ธ Use IMBENS for class-imbalanced classification with <5 lines of code:

# Train an SPE classifier
from imbens.ensemble import SelfPacedEnsembleClassifier
clf = SelfPacedEnsembleClassifier(random_state=42)
clf.fit(X_train, y_train)

# Predict with an SPE classifier
y_pred = clf.predict(X_test)

๐Ÿค— Citing IMBENS

๐Ÿป We appreciate your citation if you find our work helpful! The BibTeX entry:

@article{liu2023imbens,
  title={IMBENS: Ensemble Class-imbalanced Learning in Python},
  author={Liu, Zhining and Kang, Jian and Tong, Hanghang and Chang, Yi},
  journal={arXiv preprint arXiv:2111.12776},
  year={2023}
}

๐Ÿ‘ฏโ€โ™‚๏ธ Contribute to IMBENS

Join us and become a contributor! Please refer to the contributing guidelines.

๐Ÿ“š Table of Contents

Installation

It is recommended to use pip for installation.
Please make sure the latest version is installed to avoid potential problems:

$ pip install imbalanced-ensemble            # normal install
$ pip install --upgrade imbalanced-ensemble  # update if needed

Or you can install imbalanced-ensemble by clone this repository:

$ git clone https://github.com/ZhiningLiu1998/imbalanced-ensemble.git
$ cd imbalanced-ensemble
$ pip install .

imbalanced-ensemble requires following dependencies:

List of implemented methods

Currently (v0.1.3, 2021/06), 16 ensemble imbalanced learning methods were implemented:
(Click to jump to the document page)

Note: imbalanced-ensemble is still under development, please see API reference for the latest list.

5-min Quick Start with IMBENS

Here, we provide some quick guides to help you get started with IMBENS.
We strongly encourage users to check out the example gallery for more comprehensive usage examples, which demonstrate many advanced features of IMBENS.

A minimal working example

Taking self-paced ensemble [1] as an example, it only requires less than 10 lines of code to deploy it:

>>> from imbens.ensemble import SelfPacedEnsembleClassifier
>>> from sklearn.datasets import make_classification
>>> from sklearn.model_selection import train_test_split
>>> 
>>> X, y = make_classification(n_samples=1000, n_classes=3,
...                            n_informative=4, weights=[0.2, 0.3, 0.5],
...                            random_state=0)
>>> X_train, X_test, y_train, y_test = train_test_split(
...                            X, y, test_size=0.2, random_state=42)
>>> clf = SelfPacedEnsembleClassifier(random_state=0)
>>> clf.fit(X_train, y_train)
SelfPacedEnsembleClassifier(...)
>>> clf.predict(X_test)  
array([...])

Visualize ensemble classifiers

The imbens.visualizer sub-module provide an ImbalancedEnsembleVisualizer. It can be used to visualize the ensemble estimator(s) for further information or comparison. Please refer to visualizer documentation and examples for more details.

Fit an ImbalancedEnsembleVisualizer

from imbens.ensemble import SelfPacedEnsembleClassifier
from imbens.ensemble import RUSBoostClassifier
from imbens.ensemble import EasyEnsembleClassifier
from sklearn.tree import DecisionTreeClassifier

# Fit ensemble classifiers
init_kwargs = {'estimator': DecisionTreeClassifier()}
ensembles = {
    'spe': SelfPacedEnsembleClassifier(**init_kwargs).fit(X_train, y_train),
    'rusboost': RUSBoostClassifier(**init_kwargs).fit(X_train, y_train),
    'easyens': EasyEnsembleClassifier(**init_kwargs).fit(X_train, y_train),
}

# Fit visualizer
from imbens.visualizer import ImbalancedEnsembleVisualizer
visualizer = ImbalancedEnsembleVisualizer().fit(ensembles=ensembles)

Plot performance curves

fig, axes = visualizer.performance_lineplot()

Plot confusion matrices

fig, axes = visualizer.confusion_matrix_heatmap()

Customizing training log

All ensemble classifiers in IMBENS support customizable training logging. The training log is controlled by 3 parameters eval_datasets, eval_metrics, and training_verbose of the fit() method. Read more details in the fit documentation.

Enable auto training log

clf.fit(..., train_verbose=True)
โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”ณโ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”ณโ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”“
โ”ƒ             โ”ƒ                          โ”ƒ            Data: train             โ”ƒ
โ”ƒ #Estimators โ”ƒ    Class Distribution    โ”ƒ               Metric               โ”ƒ
โ”ƒ             โ”ƒ                          โ”ƒ  acc    balanced_acc   weighted_f1 โ”ƒ
โ”ฃโ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ•‹โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ•‹โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”ซ
โ”ƒ      1      โ”ƒ {0: 150, 1: 150, 2: 150} โ”ƒ 0.838      0.877          0.839    โ”ƒ
โ”ƒ      5      โ”ƒ {0: 150, 1: 150, 2: 150} โ”ƒ 0.924      0.949          0.924    โ”ƒ
โ”ƒ     10      โ”ƒ {0: 150, 1: 150, 2: 150} โ”ƒ 0.954      0.970          0.954    โ”ƒ
โ”ƒ     15      โ”ƒ {0: 150, 1: 150, 2: 150} โ”ƒ 0.979      0.986          0.979    โ”ƒ
โ”ƒ     20      โ”ƒ {0: 150, 1: 150, 2: 150} โ”ƒ 0.990      0.993          0.990    โ”ƒ
โ”ƒ     25      โ”ƒ {0: 150, 1: 150, 2: 150} โ”ƒ 0.994      0.996          0.994    โ”ƒ
โ”ƒ     30      โ”ƒ {0: 150, 1: 150, 2: 150} โ”ƒ 0.988      0.992          0.988    โ”ƒ
โ”ƒ     35      โ”ƒ {0: 150, 1: 150, 2: 150} โ”ƒ 0.999      0.999          0.999    โ”ƒ
โ”ƒ     40      โ”ƒ {0: 150, 1: 150, 2: 150} โ”ƒ 0.995      0.997          0.995    โ”ƒ
โ”ƒ     45      โ”ƒ {0: 150, 1: 150, 2: 150} โ”ƒ 0.995      0.997          0.995    โ”ƒ
โ”ƒ     50      โ”ƒ {0: 150, 1: 150, 2: 150} โ”ƒ 0.993      0.995          0.993    โ”ƒ
โ”ฃโ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ•‹โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ•‹โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”ซ
โ”ƒ    final    โ”ƒ {0: 150, 1: 150, 2: 150} โ”ƒ 0.993      0.995          0.993    โ”ƒ
โ”—โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”ปโ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”ปโ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”›

Customize granularity and content of the training log

clf.fit(..., 
        train_verbose={
            'granularity': 10,
            'print_distribution': False,
            'print_metrics': True,
        })
Click to view example output 
โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”ณโ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”“
โ”ƒ             โ”ƒ            Data: train             โ”ƒ
โ”ƒ #Estimators โ”ƒ               Metric               โ”ƒ
โ”ƒ             โ”ƒ  acc    balanced_acc   weighted_f1 โ”ƒ
โ”ฃโ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ•‹โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”ซ
โ”ƒ      1      โ”ƒ 0.964      0.970          0.964    โ”ƒ
โ”ƒ     10      โ”ƒ 1.000      1.000          1.000    โ”ƒ
โ”ƒ     20      โ”ƒ 1.000      1.000          1.000    โ”ƒ
โ”ƒ     30      โ”ƒ 1.000      1.000          1.000    โ”ƒ
โ”ƒ     40      โ”ƒ 1.000      1.000          1.000    โ”ƒ
โ”ƒ     50      โ”ƒ 1.000      1.000          1.000    โ”ƒ
โ”ฃโ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ•‹โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”ซ
โ”ƒ    final    โ”ƒ 1.000      1.000          1.000    โ”ƒ
โ”—โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”ปโ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”›

Add evaluation dataset(s)

  clf.fit(..., 
          eval_datasets={
              'valid': (X_valid, y_valid)
          })
Click to view example output 
โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”ณโ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”ณโ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”“
โ”ƒ             โ”ƒ            Data: train             โ”ƒ            Data: valid             โ”ƒ
โ”ƒ #Estimators โ”ƒ               Metric               โ”ƒ               Metric               โ”ƒ
โ”ƒ             โ”ƒ  acc    balanced_acc   weighted_f1 โ”ƒ  acc    balanced_acc   weighted_f1 โ”ƒ
โ”ฃโ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ•‹โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ•‹โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”ซ
โ”ƒ      1      โ”ƒ 0.939      0.961          0.940    โ”ƒ 0.935      0.933          0.936    โ”ƒ
โ”ƒ     10      โ”ƒ 1.000      1.000          1.000    โ”ƒ 0.971      0.974          0.971    โ”ƒ
โ”ƒ     20      โ”ƒ 1.000      1.000          1.000    โ”ƒ 0.982      0.981          0.982    โ”ƒ
โ”ƒ     30      โ”ƒ 1.000      1.000          1.000    โ”ƒ 0.983      0.983          0.983    โ”ƒ
โ”ƒ     40      โ”ƒ 1.000      1.000          1.000    โ”ƒ 0.983      0.982          0.983    โ”ƒ
โ”ƒ     50      โ”ƒ 1.000      1.000          1.000    โ”ƒ 0.983      0.982          0.983    โ”ƒ
โ”ฃโ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ•‹โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ•‹โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”ซ
โ”ƒ    final    โ”ƒ 1.000      1.000          1.000    โ”ƒ 0.983      0.982          0.983    โ”ƒ
โ”—โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”ปโ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”ปโ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”›

Customize evaluation metric(s)

from sklearn.metrics import accuracy_score, f1_score
clf.fit(..., 
        eval_metrics={
            'acc': (accuracy_score, {}),
            'weighted_f1': (f1_score, {'average':'weighted'}),
        })
Click to view example output 
โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”ณโ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”ณโ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”“
โ”ƒ             โ”ƒ     Data: train      โ”ƒ     Data: valid      โ”ƒ
โ”ƒ #Estimators โ”ƒ        Metric        โ”ƒ        Metric        โ”ƒ
โ”ƒ             โ”ƒ  acc    weighted_f1  โ”ƒ  acc    weighted_f1  โ”ƒ
โ”ฃโ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ•‹โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ•‹โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”ซ
โ”ƒ      1      โ”ƒ 0.942      0.961     โ”ƒ 0.919      0.936     โ”ƒ
โ”ƒ     10      โ”ƒ 1.000      1.000     โ”ƒ 0.976      0.976     โ”ƒ
โ”ƒ     20      โ”ƒ 1.000      1.000     โ”ƒ 0.977      0.977     โ”ƒ
โ”ƒ     30      โ”ƒ 1.000      1.000     โ”ƒ 0.981      0.980     โ”ƒ
โ”ƒ     40      โ”ƒ 1.000      1.000     โ”ƒ 0.980      0.979     โ”ƒ
โ”ƒ     50      โ”ƒ 1.000      1.000     โ”ƒ 0.981      0.980     โ”ƒ
โ”ฃโ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ•‹โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ•‹โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”ซ
โ”ƒ    final    โ”ƒ 1.000      1.000     โ”ƒ 0.981      0.980     โ”ƒ
โ”—โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”ปโ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”ปโ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”โ”›

About imbalanced learning

Class-imbalance (also known as the long-tail problem) is the fact that the classes are not represented equally in a classification problem, which is quite common in practice. For instance, fraud detection, prediction of rare adverse drug reactions and prediction gene families. Failure to account for the class imbalance often causes inaccurate and decreased predictive performance of many classification algorithms. Imbalanced learning aims to tackle the class imbalance problem to learn an unbiased model from imbalanced data.

For more resources on imbalanced learning, please refer to awesome-imbalanced-learning.

Acknowledgements

IMBENS was initially developed on top of imbalanced-learn, but has undergone heavy developments to implement many important imbalanced ensemble techniques. The infrastructure also underwent significant refactoring to support advanced ensemble learning features that are essential to practical usability (fine-grained training control, parallel computing, multi-class support, training logs, visualization, etc).

References

# Reference
[1] Zhining Liu, Wei Cao, Zhifeng Gao, Jiang Bian, Hechang Chen, Yi Chang, and Tie-Yan Liu. 2019. Self-paced Ensemble for Highly Imbalanced Massive Data Classification. 2020 IEEE 36th International Conference on Data Engineering (ICDE). IEEE, 2020, pp. 841-852.
[2] X.-Y. Liu, J. Wu, and Z.-H. Zhou, Exploratory undersampling for class-imbalance learning. IEEE Transactions on Systems, Man, and Cybernetics, Part B (Cybernetics), vol. 39, no. 2, pp. 539โ€“550, 2009.
[3] Chen, Chao, Andy Liaw, and Leo Breiman. โ€œUsing random forest to learn imbalanced data.โ€ University of California, Berkeley 110 (2004): 1-12.
[4] C. Seiffert, T. M. Khoshgoftaar, J. Van Hulse, and A. Napolitano, Rusboost: A hybrid approach to alleviating class imbalance. IEEE Transactions on Systems, Man, and Cybernetics-Part A: Systems and Humans, vol. 40, no. 1, pp. 185โ€“197, 2010.
[5] Maclin, R., & Opitz, D. (1997). An empirical evaluation of bagging and boosting. AAAI/IAAI, 1997, 546-551.
[6] N. V. Chawla, A. Lazarevic, L. O. Hall, and K. W. Bowyer, Smoteboost: Improving prediction of the minority class in boosting. in European conference on principles of data mining and knowledge discovery. Springer, 2003, pp. 107โ€“119
[7] S. Wang and X. Yao, Diversity analysis on imbalanced data sets by using ensemble models. in 2009 IEEE Symposium on Computational Intelligence and Data Mining. IEEE, 2009, pp. 324โ€“331.
[8] Fan, W., Stolfo, S. J., Zhang, J., & Chan, P. K. (1999, June). AdaCost: misclassification cost-sensitive boosting. In Icml (Vol. 99, pp. 97-105).
[9] Shawe-Taylor, G. K. J., & Karakoulas, G. (1999). Optimizing classifiers for imbalanced training sets. Advances in neural information processing systems, 11(11), 253.
[10] Viola, P., & Jones, M. (2001). Fast and robust classification using asymmetric adaboost and a detector cascade. Advances in Neural Information Processing System, 14.
[11] Freund, Y., & Schapire, R. E. (1997). A decision-theoretic generalization of on-line learning and an application to boosting. Journal of computer and system sciences, 55(1), 119-139.
[12] Breiman, L. (1996). Bagging predictors. Machine learning, 24(2), 123-140.
[13] Guillaume Lemaรฎtre, Fernando Nogueira, and Christos K. Aridas. Imbalanced-learn: A python toolbox to tackle the curse of imbalanced datasets in machine learning. Journal of Machine Learning Research, 18(17):1โ€“5, 2017.

Related Projects

Check out Zhining's other open-source projects!


Imbalanced Learning [Awesome]
GitHub stars
Machine Learning [Awesome]
GitHub stars
Self-paced Ensemble [ICDE]
GitHub stars
Meta-Sampler [NeurIPS]
GitHub stars

Contributors โœจ

Thanks goes to these wonderful people (emoji key):

Zhining Liu
Zhining Liu
๐Ÿ’ป ๐Ÿค” ๐Ÿšง ๐Ÿ› ๐Ÿ“–		 leaphan
leaphan
๐Ÿ›		 hannanhtang
hannanhtang
๐Ÿ›		 H.J.Ren
H.J.Ren
๐Ÿ›		 Marc Skov Madsen
Marc Skov Madsen
๐Ÿ›

This project follows the all-contributors specification. Contributions of any kind welcome!