Collect feedback on MLflow logging

[williambdean]

Handful of items are logged now. Would like to know what else would be useful? Thinks like:
General PyMC

• Total samples

MMM

• saturation curves
• posterior plots

Doesnt need to be in the autologging per say

[williambdean]

Many of the params should be logged before the pm.sample in the autologging. This way, the UI gets information about the run before a long function call happens

[williambdean]

Parameters will act like strings. So storing number of sample would not allow for comparisons like: params.posterior_samples >= 2000

Thus, it might make sense to store as a metric for better experience in the MLflow UI. Though it might be handled differently in tools like DataBricks

Reference: https://mlflow.org/docs/latest/search-runs.html#searching-by-params