Blog Logo

2019-03-03T22:40:32.169Z ~ 2 min read

Generative model for text mock-data


tl;dr pip install textgen :rocket:

Recently went to Stockholm AI study group; where Judith presented a talk on generative models.

Generative model

A generative model is basically a overfitted model for trying to describe the underlying data and being able to generate predictions based on it, by providing either a encoding and a decoder or a distribution to sample from.

This can be used for generating mock-data for any type of text-based entity or column (bare in mind that it should not be cruicial to your organization ofc). Having a set of known text-based features that gets presented in demos or at any upstart of a project can be generated by a generative model on a learned distribution that serves as the dummy data creator.

As with every python implementation, there is usually always a python package for that, as with minimaxir/textgen.

However, it took some time to get everything setup so I prepared for myself some handy little snippets and docker file for getting everything setup, as had to get randomly generated data for a multitude of various sources and found that I would provide a simple snippet to interact with this wonderful package.

If you have not used keras/deep learning but want to get it setup and use it. In the repo there is documentation on how to get started eleijonmarck/data-generator

  1. Get the data to sample-datasets, with a column_name and the text data you want to generate.
  2. Train the generator using the textgen/ or make example-train

Headshot of Moi

Hi, I'm Eric. I'm a software engineer and data scientist based in Lisbon. You can follow me on Twitter, see some of my work on GitHub,