Why do timestamps matter? Traditionally, they were used by Darwin to determine if the problem was time-series or not. However, in most model building cases, they're completely ignored. If a dataset has a date/time stamp, but is not a time-series problem, certain ways in which models are developed end up being less accurate. To get around that, we directly ask a user in our latest workflow if it is a Time Series problem. We can then use their answer to that question to change how we shuffle, turn on recurrency, and change imputation. So for these, dropping a date/time ends up not affecting the problem, as we still treat the model as time series based on the user answer.
Could the time stamp be useful for something else, like ordering the data? That is absolutely a viable concern, but something not addressed by Darwin. Darwin assumes that all date/time data is already presented in order and does not re-shuffle the data based on the time stamp.
Could I use this to create new features? Yes, you absolutely could, and this is something that is on our roadmap to support. However, we do not yet support it.
What if I wanted to use the date/time stamp for grouping data? We are adding this capability for November, so that concern will be addressed.
Could the time stamp be used as a categorical feature? In this case, maybe. We will plan on addressing this in a future release, as Darwin currently limits the number of unique categories it considers in a categorical variable.