Employing point-in-time lookups for time series feature tables
Time series feature tables are any table in Unity Catalog with a TIMESERIES
primary key. These tables are eligible for point-in-time lookups, which is a mechanism for looking up the correct feature values. Before training_sets
, coming in Chapter 6, we often joined tables to connect training rows with their feature values. However, a fine-grained event timestamp is not ideal for joining. This led to rounding the timestamps to minutes, hours, or even days. Depending on the use case, this method may or may not work. For example, joining on TransactionTimestamp
in Figure 5.3 is not realistic in a standard join so one might create TransactionMinute
or TransactionHour
.
TransactionTimestamp |
TransactionMinute |
TransactionHour |
2023-09-03 19:23:09.765676 |
2023-09-03 19:23:00 ... |