Estimating the branching factor

瀏覽次數:64 次
跳到第一則未讀訊息

Rob

未讀,
2013年1月3日 清晨5:34:082013/1/3
收件者:google-app-en...@googlegroups.com
Hello

I would like to know if there should be an estimation of the branching factor to pursuit a certain number of nodes. How can someone know if the tree/number of nodes is big/small enough?

Thank you.

Bartholomew Furrow

未讀,
2013年1月3日 晚上8:57:182013/1/3
收件者:google-app-en...@googlegroups.com
If you aren't going to get many write QPS on the ranker, it probably doesn't matter.

More levels means fewer transactional collisions, but also more nodes per lookup.

More nodes per lookup, as long as the number is reasonably small (<20?), isn't really a problem for SetScore and FindRank, but it is for FindScore, which can end up doing serial lookups proportional to the depth of the tree.

Fewer transactional collisions is important if you're going to end up with a lot of writes (SetScore).

So as with all things, it's a tradeoff.  We found that a depth of 5 and a branching factor of 101 (scores 0-9999, penalty times 0-999999 seconds) worked pretty well for Code Jam, which has a lot of reads per write.

I hope that helps!
Bartholomew



--
You received this message because you are subscribed to the Google Groups "Google App Engine Ranklist" group.
To view this discussion on the web visit https://groups.google.com/d/msg/google-app-engine-ranklist/-/cEw11vOVpwEJ.
To post to this group, send email to google-app-en...@googlegroups.com.
To unsubscribe from this group, send email to google-app-engine-r...@googlegroups.com.
For more options, visit this group at http://groups.google.com/group/google-app-engine-ranklist?hl=en.

回覆所有人
回覆作者
轉寄
0 則新訊息