Package index • themis

Over-sampling

Over-sampling is the act of synthetically generating observations for the minority classes, either by random replication or by more advanced methods that synthesize new observations.

Random replication

step_upsample(): Up-Sample a Data Set Based on a Factor Variable

SMOTE family

step_smote(): Apply SMOTE Algorithm
step_bsmote(): Apply Borderline-SMOTE Algorithm
step_svmsmote(): Apply SVM-SMOTE Algorithm
step_kmeans_smote(): Apply KMeans-SMOTE Algorithm
step_smotenc(): Apply SMOTENC Algorithm
step_smoten(): Apply SMOTEN Algorithm

Adaptive

step_adasyn(): Apply Adaptive Synthetic Algorithm

Smoothed bootstrap

step_rose(): Apply ROSE Algorithm

Under-sampling

Under-sampling is the act of removing observations from the majority classes. Most under-samplers here use prototype selection, keeping a subset of the real rows. step_cluster_centroids() instead uses prototype generation, creating new representative rows.

Random

step_downsample(): Down-Sample a Data Set Based on a Factor Variable

Distance / near-boundary

step_nearmiss(): Remove Points Near Other Classes
step_tomek(): Remove Tomek's Links
step_cnn(): Condensed Nearest Neighbors
step_oss(): One-Sided Selection

Neighborhood cleaning

step_enn(): Edited Nearest Neighbors
step_ncl(): Neighborhood Cleaning Rule

Hardness

step_instance_hardness(): Remove hard to classify points

Prototype generation

step_cluster_centroids(): Under-Sampling by Cluster Centroids

Regression

Resampling an imbalanced numeric outcome rather than a class.

step_smogn(): Apply SMOGN Algorithm

Methods

Some of the methods implemented in this package as steps are also available as their own functions.

smote(): SMOTE Algorithm
smogn(): SMOGN Algorithm
smotenc(): SMOTENC Algorithm
smoten(): SMOTEN Algorithm
bsmote(): Borderline-SMOTE Algorithm
svmsmote(): SVM-SMOTE Algorithm
kmeans_smote(): KMeans-SMOTE Algorithm
adasyn(): Adaptive Synthetic Algorithm
rose(): ROSE Algorithm
cluster_centroids(): ClusterCentroids Algorithm
cnn(): Condensed Nearest Neighbors
enn(): Edited Nearest Neighbors
instance_hardness(): Remove hard to classify points
ncl(): Neighborhood Cleaning Rule
nearmiss(): Remove Points Near Other Classes
oss(): One-Sided Selection
tomek(): Remove Tomek's Links

Datasets

circle_example: Synthetic Dataset With a Circle

Developer tools

required_pkgs(<step_adasyn>) required_pkgs(<step_bsmote>) required_pkgs(<step_cluster_centroids>) required_pkgs(<step_cnn>) required_pkgs(<step_downsample>) required_pkgs(<step_enn>) required_pkgs(<step_instance_hardness>) required_pkgs(<step_kmeans_smote>) required_pkgs(<step_ncl>) required_pkgs(<step_nearmiss>) required_pkgs(<step_oss>) required_pkgs(<step_rose>) required_pkgs(<step_smogn>) required_pkgs(<step_smote>) required_pkgs(<step_smoten>) required_pkgs(<step_smotenc>) required_pkgs(<step_svmsmote>) required_pkgs(<step_tomek>) required_pkgs(<step_upsample>): S3 methods for tracking which additional packages are needed for steps.