Skip to content
Accelerating Extreme Classification via Adaptive Feature Agglomeration
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.
Tools/c++ Add files via upload May 20, 2019
sandbox/bibtex Add files via upload May 20, 2019 Update Jun 16, 2019 Update May 20, 2019



Accelerating Extreme Classification via Adaptive Feature Agglomeration

Running DEFRAG

DEFARG is executed in two steps:

  • defrag_clustering: This computes a grouping of features.
  • defrag_agglomeration: This agglomerates the features based on groupings obtained from previous step.

Please refer to for more information on how to use DEFRAG.

Feature and label files should be formatted as expected by Parabel.


Following parameters can be tuned in DEFRAG


fr = param.feature_representation : Use feture repersentation X or XY, default 1 (X).
cml = param.cluster_maxleaf : Maximum number of features in a leaf node of DEFRAG tree, default 8.
cls = param.cluster_label_sample : Percentage of labels used for clustering, default 5.
cds = param.cluster_data_sample : Percentage of data points used for clustering, default 20.


avg = param.avg : Average out non-zero entries while agglomeration, default 0"<<endl;


The code is adapted and subsequently modified from the source code provided by the authors of Parabel: Partitioned Label Trees for Extreme Classification with Application to Dynamic Search Advertising.

You can’t perform that action at this time.