/synthetic-data-generation

Synthetic data generation approaches for tabular and time series data using LLM, Copulas and CTGAN

Primary LanguageJupyter Notebook

synthetic-data-generation

We will covering Synthetic data generation using :

  1. LLM
  2. Synthetic data vault
  1. LLM (Pre-trained Large Language model) : In this we are going leverage the distilGPT2 model and be_great package to generate the synthetic tabular data.

GReaT stands for Generation of Realistic Tabular data with pretrained Transformer-based language models Following is claim of the authors and we can see that it is performing better than CouplaGAN and CTGAN

Alt text

  1. Synthetic data vault: This library provides option to generate synthetic data using mutiple algorithm.

    2.1 GaussianCoupla: Helps modeling complex relationships between variables

    2.2 CTGAN (Conditional Transform Generative Adverserial Network)

    2.3 PAR (Probabilistic Autoregressive for Time-Series or Sequence data)