/datasets4vw

Multiclass and multilabel datasets in Vowpal Wabbit format

Primary LanguageShell

datasets4vw

Scripts to download and convert different datasets to Vowpal Wabbit format. Repository contains scripts for the following datasets:

Usage: bash ./<repo>/get_<dataset-name>.sh

xml_repo (multilabel):

  • Amazon-3M (amazon-3M)
  • Amazon-670K (amazon)
  • AmazonCat-13K (amazonCat)
  • AmazonCat-14K (amazonCat-14K)
  • Bibtex (bibtex)
  • Delicious (delicious)
  • Delicious-200K (deliciousLarge)
  • EURLex-4K (eurlex)
  • Mediamill (mediamill)
  • RCV1-2K (rcv1x)
  • Wiki10-31K (wiki10)
  • WikiLSHTC-325K (wikiLSHTC)
  • Wikipedia-500K (WikipediaLarge-500K)

pd_sparse(_multiclass):

  • aloi.bin
  • Dmoz
  • imageNet
  • LSHTC1
  • sector

pd_sparse_multilabel:

  • Eur-Lex
  • rcv1_regions
  • bibtex
  • LSHTCwiki

Mac users need to install gnu version of sed:

brew install gnu-sed