Thanks Mary, though I am still not very clear how that logtf is calculated :) About the max weight, when I said I can set it up to <=67. I meant when I tried to set to, for example, 68, the "word query" setting page cannot show the element name I just configured, it shows nothing (I am using 7.0-2.1). So, in this case, I think it is not allowed. But, you are right, the score is scaled down. John
On Thu, Apr 24, 2014 at 9:28 AM, Mary Holstege <[email protected]>wrote: > On Wed, 23 Apr 2014 17:08:21 -0700, John Zhong <[email protected]> > wrote: > > Hi, >> >> I have few questions about the cts:relevance-info result, please see >> below, >> when doing the search, I use the default scoring method, according to the >> document, it says formula: log(term frequency) * (inverse document >> frequency). My questions are: >> >> 1, how the logtf = 13 is calculated? In my document, it has only two hits >> for word "Test", in this case, I think tf = term frequency = 2 (?), then >> log(2)=0.3 ? And please note the two hits are from two different elements >> which have different weight setting in the database>Word Query setting, >> because I want some elements returned higher score. >> > > We use integer arithmetic for all scoring calculations, and tables for > much of it, so the logtf values are scaled, capped at minimum and > maximum values, and bucketed. This is why you see otherwise mysterious > constants like 8 and 256 in the score calculations. The minimum (non-zero) > logtf value is 8. > > > 2, what is the max weight can we set? I read some document, it says it is >> 16, but also mentions "super weight +64", I tried to set a higher number, >> I >> found I can set it up to <=67. >> > > We won't stop you passing other values. If you put in something more than > 64, > we'll scale it down to the range. We will also not stop you putting in > values between 16 and 64, but the intended usage is to keep a clear > separation > between normal weights in the +/-16 range and superboosting which is well > above that (64). > > //Mary >
_______________________________________________ General mailing list [email protected] http://developer.marklogic.com/mailman/listinfo/general
