A Python Toolbox for Machine Learning Model Combination
Project description
Deployment & Documentation & Stats
Build Status & Coverage & Maintainability & License
combo is a Python toolbox for combining or aggregating ML models and scores for various tasks, including classification, clustering, anomaly detection, and raw score. It has been widely used in data science competitions and real-world tasks, such as Kaggle.
Model and score combination can be regarded as a subtask of ensemble learning, but is often beyond the scope of ensemble learning. For instance, averaging the results of multiple runs of a ML model is deemed as a reliable way of eliminating the randomness for better stability. See figure below for some popular combination approaches.
combo is featured for:
Unified APIs, detailed documentation, and interactive examples across various algorithms.
Advanced models, including dynamic classifier/ensemble selection and LSCP.
Broad applications for classification, clustering, anomaly detection, and raw score.
Comprehensive coverage for supervised, unsupervised, and semi-supervised scenarios.
Optimized performance with JIT and parallelization when possible, using numba and joblib.
API Demo:
from combo.models.stacking import Stacking # base classifiers classifiers = [DecisionTreeClassifier(), LogisticRegression(), KNeighborsClassifier(), RandomForestClassifier(), GradientBoostingClassifier()] clf = Stacking(base_clfs=classifiers) # initialize a Stacking model clf.fit(X_train) # predict on unseen data y_test_labels = clf.predict(X_test) # label prediction y_test_proba = clf.predict_proba(X_test) # probability prediction
Table of Contents:
Installation
It is recommended to use pip for installation. Please make sure the latest version is installed, as combo is updated frequently:
pip install combo # normal install
pip install --upgrade combo # or update if needed
pip install --pre combo # or include pre-release version for new features
Alternatively, you could clone and run setup.py file:
git clone https://github.com/yzhao062/combo.git
cd combo
pip install .
Required Dependencies:
Python 3.5, 3.6, or 3.7
joblib
matplotlib
numpy>=1.13
numba>=0.35
scipy>=0.19.1
scikit_learn>=0.19.1
API Cheatsheet & Reference
Full API Reference: (https://pycombo.readthedocs.io/en/latest/api.html). API cheatsheet for most of the models:
fit(X): Fit an estimator.
predict(X): Predict on a particular sample once the estimator is fitted.
predict_proba(X): Predict the probability of a sample belonging to each class. Only applicable for classification tasks.
Proposed Algorithms
combo groups combination frameworks by tasks.
For most of the tasks, the following combination methods for raw scores are feasible [7]:
Averaging & Weighted Averaging & Median
Maximization
Majority Vote & Weighted Majority Vote
Median
Some of the methods are tasks specific:
Classifier combination: combine multiple supervised classifiers together for training and prediction
SimpleClassifierAggregator: combining classifiers by (i) (weighted) average (ii) maximization (iii) median and (iv) (weighted) majority vote
Dynamic Classifier Selection & Dynamic Ensemble Selection [3] (work-in-progress)
Stacking (meta ensembling): build an additional classifier to learn base estimator weights [2]
Cluster combination: combine and align unsupervised clustering results
Clusterer Ensemble [6]
Anomaly detection: combine unsupervised (and supervised) outlier detectors
SimpleDetectorCombination: combining outlier score results by (i) (weighted) average (ii) maximization (iii) median and (iv) (weighted) majority vote
Average of Maximum (AOM) [1]
Maximum of Average (MOA) [1]
Thresholding
Locally Selective Combination (LSCP) [4]
XGBOD: a semi-supervised combination framework for outlier detection [5]
Quick Start for Classifier Combination
“examples/classifier_comb_example.py” demonstrates the basic API of predicting with multiple classifiers. It is noted that the API across all other algorithms are consistent/similar.
Initialize a group of classifiers as base estimators
# initialize a group of classifiers classifiers = [DecisionTreeClassifier(), LogisticRegression(), KNeighborsClassifier(), RandomForestClassifier(), GradientBoostingClassifier()]
Initialize, fit, predict, and evaluate with a simple aggregator (average)
from combo.models.classifier_comb import SimpleClassifierAggregator clf = SimpleClassifierAggregator(classifiers, method='average') clf.fit(X_train, y_train) y_test_predicted = clf.predict(X_test) evaluate_print('Combination by avg |', y_test, y_test_predicted)
See a sample output of classifier_comb_example.py
Decision Tree | Accuracy:0.9386, ROC:0.9383, F1:0.9521 Logistic Regression | Accuracy:0.9649, ROC:0.9615, F1:0.973 K Neighbors | Accuracy:0.9561, ROC:0.9519, F1:0.9662 Gradient Boosting | Accuracy:0.9605, ROC:0.9524, F1:0.9699 Random Forest | Accuracy:0.9605, ROC:0.961, F1:0.9693 Combination by avg | Accuracy:0.9693, ROC:0.9677, F1:0.9763 Combination by w_avg | Accuracy:0.9781, ROC:0.9716, F1:0.9833 Combination by max | Accuracy:0.9518, ROC:0.9312, F1:0.9642 Combination by w_vote| Accuracy:0.9649, ROC:0.9644, F1:0.9728 Combination by median| Accuracy:0.9693, ROC:0.9677, F1:0.9763
Quick Start for Clustering Combination
“examples/cluster_comb_example.py” demonstrates the basic API of combining multiple base clustering estimators.
Initialize a group of clustering methods as base estimators
from combo.models.cluster_comb import ClustererEnsemble # Initialize a set of estimators estimators = [KMeans(n_clusters=n_clusters), MiniBatchKMeans(n_clusters=n_clusters), AgglomerativeClustering(n_clusters=n_clusters)]
Initialize an Clusterer Ensemble class and fit the model
# combine by Clusterer Ensemble clf = ClustererEnsemble(estimators, n_clusters=n_clusters) clf.fit(X)
Get the aligned results
# generate the labels on X aligned_labels = clf.aligned_labels_ predicted_labels = clf.labels_
An Example of Stacking
“examples/stacking_example.py” demonstrates the basic API of stacking (meta ensembling).
Initialize a group of classifiers as base estimators
# initialize a group of classifiers classifiers = [DecisionTreeClassifier(), LogisticRegression(), KNeighborsClassifier(), RandomForestClassifier(), GradientBoostingClassifier()]
Initialize, fit, predict, and evaluate with Stacking
from combo.models.stacking import Stacking clf = Stacking(base_clfs=classifiers, n_folds=4, shuffle_data=False, keep_original=True, use_proba=False, random_state=random_state) clf.fit(X_train, y_train) y_test_predict = clf.predict(X_test) evaluate_print('Stacking | ', y_test, y_test_predict)
See a sample output of stacking_example.py
Decision Tree | Accuracy:0.9386, ROC:0.9383, F1:0.9521 Logistic Regression | Accuracy:0.9649, ROC:0.9615, F1:0.973 K Neighbors | Accuracy:0.9561, ROC:0.9519, F1:0.9662 Gradient Boosting | Accuracy:0.9605, ROC:0.9524, F1:0.9699 Random Forest | Accuracy:0.9605, ROC:0.961, F1:0.9693 Stacking | Accuracy:0.9868, ROC:0.9841, F1:0.9899
Development Status
combo is currently under development as of July 24, 2019. A concrete plan has been laid out and will be implemented in the next few months.
Similar to other libraries built by us, e.g., Python Outlier Detection Toolbox (pyod), combo is also targeted to be published in Journal of Machine Learning Research (JMLR), open-source software track. A demo paper to AAAI or IJCAI may be submitted soon for progress update.
Watch & Star to get the latest update! Also feel free to send me an email (zhaoy@cmu.edu) for suggestions and ideas.
Reference
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.