- From: Ruben Verborgh <ruben.verborgh@ugent.be>
- Date: Wed, 27 May 2015 15:20:37 +0200
- To: Miel Vander Sande <miel.vandersande@ugent.be>
- Cc: public-hydra@w3.org
Hi Miel, > In context of reading on detecting dataset changes, I was wondering if we ever discussed the stability of count estimates across TPF requests? As far as I'm concerned, no stability is required. Neither by the VoID spec (which we cannot change), neither by the TPF spec (which we can change if desired). The higher the accuracy, the better, but we do not place a bound on this, nor do we require the server to place a bound (such as “close to previous estimate”). The only hard requirements at the moment are: – if no triples match the selector, the estimate MUST be zero. – If one or more triples match the selector, the estimate MUST NOT be zero. Could there be issues when there is no stability? > In other words, presuming caching has expired already, will a request to the same TPF today and tomorrow guarantee the same estimate if the data hasn't changed? Not necessarily. For instance, the estimation algorithm could improve. > I found no mention of this in the spec: http://www.hydra-cg.com/spec/latest/triple-pattern-fragments/#metadata. Should we explicitly mention that changes are allowed? I think it is sufficient to not specify it, but I'm open to suggestions. Best, Ruben
Received on Wednesday, 27 May 2015 13:21:10 UTC