Should we apply `NewRowSynthesis` by default?

[npatki]

Version: 0.8.0 (in developement)

Problem Description

Currently, we are applying the SDMetrics NewRowSynthesis by default in the benchmark_single_table script. The motivation was to capture whether new synthetic data is being created at all -- or whether the rows are being re-used as in DataIdentity.

But in practice, the NewRowSynthesis metric may not be too robust. It may error out on a large # of columns, and leading to generally longer benchmarking runs.

Expected behavior

We should consider the behavior of the default NewRowSynthesis metric that we apply:

1. We could disable it. That is, by default set sdmetrics=None
2. We could fix the underlying issues with it in the SDMetrics library. Perhaps that can achieved by subsetting or some other means.