erwv2cdo

erwv2cdo: Extended Revised Word Vectors Training Program v2 with Compass-Related Features for Diachronic Studies and Low-Frequency Optimizations

https://codeberg.org/aroussel/erwv2cdo

Usage

TODO

Requirements

Requires GNU Make and a C compiler, either gcc or clang. Tested on Debian 12 and 13, but may also compile and run in other environments.

Extensions

• Python module for scripting the C code

• Model saving/loading

  • New model format stores all parameters, so that, among other things, training can be resumed.
  • Original Word2Vec format also supported.

• Freezing of selected parameters during training

  • Allows the use of the CADE approach

• Multi-prototype learning according to AdaGram

  • Extended for use with negative sampling
  • Extended for use with CBOW
  • Including a new sense merging feature

• Optional low-frequency optimizations inspired by Nonce2Vec

Origin of the Name

‘erwv2cdo’ is a random permutation of ‘word2vec’, on which this program is based.

License

This program is free software, made available under the terms of the GNU General Public License v3.0 or later.

This code is based on word2vec, originally hosted at https://code.google.com/p/word2vec/, now available from https://github.com/tmikolov/word2vec.

The complete text of the relevant licenses can be found in the LICENSES/ directory.