Placeholder issue for dataset as configs as mentioned in #15
Also related to #45

Why?

1. It will be easier to have reproducible datasets. For example, I can share a config with another person, she/he can create the same dataset using just the config file.
2. No need to write lengthy python scripts/notebooks to generate data. The most frequent use case for this package is probably generate a diverse dataset with all sorts of models. Writing a script from scratch to generate a dataset is tedious.

Proof of Concept

Let's do a small pilot experiment to show if we can achieve the following.

• Can we fully define a dataset with a pinned package version and a config file, e.g., yaml?

Experiment: #47