Supported Machine Learning

Supported Scikit-learn

Below is the list of scikit-learn classes and functions that Bodo supports natively inside JIT functions. This list will expand regularly as we add support for more APIs. Optional arguments are not supported unless specified.

Linear Classifiers

sklearn.linear_model.LogisticRegression ^*

This class provides logistic regression classifier.

Methods:

• sklearn.linear_model.LogisticRegression.fit

• sklearn.linear_model.LogisticRegression.predict

• sklearn.linear_model.LogisticRegression.score

Attributes:

• sklearn.linear_model.LogisticRegression.coef_

sklearn.linear_model.SGDClassifier

This class provides linear classification models with SGD optimization which allows distributed large-scale learning.

SGDClassifier(loss='hinge') is equivalent to SVM linear classifer.

SGDClassifier(loss='log') is equivalent to logistic regression classifer.

• Supported loss functions hinge and log.

• early_stopping is not supported yet.

Methods:

• sklearn.linear_model.SGDClassifier.fit

• sklearn.linear_model.SGDClassifier.predict

• sklearn.linear_model.SGDClassifier.score

sklearn.svm.LinearSVC

This class provides Linear Support Vector Classification.

Methods:

• sklearn.svm.LinearSVC.fit

• sklearn.svm.LinearSVC.predict

• sklearn.svm.LinearSVC.score

Attributes:

• sklearn.linear_model.SGDClassifier.coef_

Linear Regressors

sklearn.linear_model.LinearRegression ^*

This class provides linear regression support. Note: Multilabel targets are not currently supported.

Methods:

• sklearn.linear_model.LinearRegression.fit

• sklearn.linear_model.LinearRegression.predict

• sklearn.linear_model.LinearRegression.score

Attributes:

• sklearn.linear_model.LinearRegression.coef_

sklearn.linear_model.Ridge ^*

This class provides ridge regression support.

Methods:

• sklearn.linear_model.Ridge.fit

• sklearn.linear_model.Ridge.predict

• sklearn.linear_model.Ridge.score

Attributes:

• sklearn.linear_model.Ridge.coef_

sklearn.linear_model.SGDRegressor

This class provides linear regression models with SGD optimization which allows distributed large-scale learning.

SGDRegressor(loss='squared_loss', penalty='None') is equivalent to linear regression.

SGDRegressor(loss='squared_loss', penalty='l2') is equivalent to Ridge regression.

SGDRegressor(loss='squared_loss', penalty='l1') is equivalent to Lasso regression.

• Supported loss function is squared_loss

• early_stopping is not supported yet.

Methods:

• sklearn.linear_model.SGDRegressor.fit

• sklearn.linear_model.SGDRegressor.predict

• sklearn.linear_model.SGDRegressor.score

sklearn.linear_model.Lasso ^*

This class provides Lasso regression support.

Methods:

• sklearn.linear_model.Lasso.fit

• sklearn.linear_model.Lasso.predict

• sklearn.linear_model.Lasso.score

* To enable distributed training across multiple nodes, Bodo uses Stochastic Gradient Descent (SGD) to train these model types. This provides a model that has similar performance as the corresponding sequential version.

To achieve that, it is highly recommended to scale your data using StandardScaler before training and/or testing the model.

See scikit-learn for more tips on how to tune model parameters for SGD here.

Clustering

sklearn.cluster.KMeans

This class provides K-Means clustering models which allows distributed large-scale unsupervised learning.

Methods:

• sklearn.cluster.KMeans.fit

• sklearn.cluster.KMeans.predict

• sklearn.cluster.KMeans.score

• sklearn.cluster.KMeans.transform

Ensemble Methods

sklearn.ensemble.RandomForestClassifier

This class provides Random Forest Classifier, an ensemble learning model, for distributed large-scale learning.

• random_state value is ignored when running on multi-node.

Methods:

• sklearn.ensemble.RandomForestClassifier.fit

• sklearn.ensemble.RandomForestClassifier.predict

• sklearn.ensemble.RandomForestClassifier.score

sklearn.ensemble.RandomForestRegressor

This class provides Random Forest Regressor, an ensemble learning model, for distributed large-scale learning.

• random_state value is ignored when running on multi-node.

Methods:

• sklearn.ensemble.RandomForestRegressor.fit

• sklearn.ensemble.RandomForestRegressor.predict

• sklearn.ensemble.RandomForestRegressor.score

Naive Bayes

sklearn.naive_bayes.MultinomialNB

This class provides Naive Bayes classifier for multinomial models with distributed large-scale learning.

Methods:

• sklearn.naive_bayes.MultinomialNB.fit

• sklearn.naive_bayes.MultinomialNB.predict

• sklearn.naive_bayes.MultinomialNB.score

Classification metrics

• sklearn.metrics.accuracy_score()

• sklearn.metrics.f1_score()

• sklearn.metrics.precision_score()

• sklearn.metrics.recall_score()

Regression metrics

• sklearn.metrics.mean_absolute_error()

• sklearn.metrics.mean_squared_error()

• sklearn.metrics.r2_score()

Data Preprocessing

sklearn.preprocessing.StandardScaler

This class provides Standard Scaler support to center your data and to scale it to achieve unit variance.

Methods:

• sklearn.preprocessing.StandardScaler.fit

• sklearn.preprocessing.StandardScaler.transform

• sklearn.preprocessing.StandardScaler.inverse_transform

sklearn.preprocessing.MinMaxScaler

This class provides MinMax Scaler support to scale your data based on the range of its features.

Methods:

• sklearn.preprocessing.MinMaxScaler.fit

• sklearn.preprocessing.MinMaxScaler.transform

• sklearn.preprocessing.MinMaxScaler.inverse_transform

sklearn.preprocessing.LabelEncoder

This class provides LabelEncoder support to encode target labels (y) with values between 0 and n-classes-1.

Methods:

• sklearn.preprocessing.LabelEncoder.fit

• sklearn.preprocessing.LabelEncoder.transform

• sklearn.preprocessing.LabelEncoder.fit_transform

Feature Extraction

sklearn.feature_extraction.text.HashingVectorizer

This class provides HashingVectorizer support to convert a collection of text documents to a matrix of token occurrences.

Methods:

• sklearn.feature_extraction.text.HashingVectorizer.fit_transform

sklearn.feature_extraction.text.CountVectorizer

This class provides CountVectorizer support to convert a collection of text documents to a matrix of token counts.

Methods:

• sklearn.feature_extraction.text.CountVectorizer.fit_transform

• sklearn.feature_extraction.text.CountVectorizer.get_feature_names

Model Selection

• sklearn.model_selection.train_test_split()

  • Currently it only supports two inputs of type numpy arrays and/or pandas dataframes.

  • Arguments train_size and test_size accept float between 0.0 and 1.0 or None only.

  • Arguments random_state and shuffle are supported.

  • Argument stratify is not supported yet.

Supported XGBoost

Below is the list of XGBoost (using the Scikit-Learn-like API) classes and functions that Bodo supports natively inside JIT functions. This list will expand regularly as we add support for more APIs.

XGBClassifier

xgboost.XGBClassifier

This class provides implementation of the scikit-learn API for XGBoost classification with distributed large-scale learning.

Methods:

• xgboost.XGBClassifier.fit

• xgboost.XGBClassifier.predict

XGBRegressor

xgboost.XGBRegressor

This class provides implementation of the scikit-learn API for XGBoost regression with distributed large-scale learning.

Methods:

• xgboost.XGBRegressor.fit

• xgboost.XGBRegressor.predict

©2021, Bodo, Inc..