Repeated sampling in a single estimator

[ak-gupta]

Description

We should add an experiment that examines the effect of repeated sampling with a single estimator. Right now, we have two major methods for bayte:

• Encode the categorical values as the mean of the posterior distribution,
• Encode the categorical value as a sample from the posterior, repeating x number of times to create x training datasets.

Since it's less likely a user will find an ensemble model useful, let's try this: encode a categorical value as x columns, each of which representing a sample from the posterior distribution.

We should also add native categorical handling from XGboost/LightGBM as a comparison point.