Likelihood of the generated data regarding the real data

[celsofranssa]

What metrics could we apply to measure the likelihood of the generated data regarding the real data?

[19956406179]

I also want to know, if I just: pip install ctgan, and then write a piece of code to use CTGAN to generate synthetic data for my dataset, how can I evaluate whether the generated data is good or bad

[npatki]

Hi @celsofranssa and @19956406179, great to see interest in CTGAN.

I would recommend looking into the SDMetrics library for anything metrics-related. There is a Quality Report you can run for a general estimate of data quality, visualization utilities, and other speciality metrics such as Data likelihood.

BTW -- while you are welcome to try using CTGAN as a standalone library, we actually recommend using it via the SDV library instead. Within SDV, the CTGANSynthesizer is a wrapper around this one, but it allows additional features such as data pre-processing, as well as more convenient visualizations, metrics, etc.