How are people handling timeseries analytics with ...
# general
p
How are people handling timeseries analytics with pinot? Looking for things like zero filing with last value (step interpolation), summing across interpolated series to get a total by time bucket, get last and first value in a time bucket, moving averages etc. Guess we can go over to python etc and implement but it would be nice for some native analytic functions.
g
I’m not well placed on this one but i was experimenting a little with Apache third eye https://readthedocs.org/projects/thirdeye/downloads/pdf/latest/ and there are some percentage based filtering and some based around holt winters
k
Most end up implementing udf to achieve that.. if you think some of these are useful beyond your use case, please file issue
p
Looks like presto has a pretty broad set of aggregation and window functions. https://prestodb.io/docs/current/functions/aggregate.html In pinot can we implement aggregates as udfs?
k
Yes
p
Nice I will see if I can build earliest / latest operators
Couldn't find an example udf which works on an aggregation. Looking for something like
Copy code
select product, groovy('{"returnType":"DOUBLE","isSingleValue":true}', 'arg[0]', myCol) as First
from table
group by product