This is an absolute, not relative. measure of error, so there's not a default 
epsilon that makes sense. I simply want to divide by this below to normalize 
the error of each point to this "average" error, which is fine if it's tiny. 

[ Full content available at: https://github.com/apache/beam/pull/6375 ]
This message was relayed via gitbox.apache.org for [email protected]

Reply via email to