What does TOP mean ?

[Douglas Tody]

The algorithm used to compute the score for a dataset is up to the
data center and service, but can include both a best-match term and a
quality term (clearly best-match would be more important for an
individual query).

The main difference between MAXREC and TOP is scoring and sort by score.
Usually if TOP is specified MAXREC would not matter as it is likely to
be much larger, but MAXREC if specified should still be in effect and
could result in an overflow indication.

A service could treat TOP the same as MAXREC (all records having the
same score), however computing a useful scoring would be preferable.

Finally, Google made billions developing just such an algorithm!  It is
what made their search service stand out from the others, in the early
days.

 	- Doug


On Mon, 27 Apr 2015, Markus Demleitner wrote:

> Hi Arnold,
>
> On Mon, Apr 27, 2015 at 09:43:46AM -0400, Arnold Rots wrote:
>> Yes, you need an observatory-specific model to calculate a score.
>> By its very nature, therefore, it will only be a relative score.
>> Example:
>> For Chandra we implemented a score that is based on instrument, exposure
>> time, and off-axis angle
>> (PSF degrades with increasing off-axis angle).
>
> Well, yes, that could be part of it, but the way I understand
> things --
>
>  [...] the general idea is that the better a candidate dataset
>  matches the query, the higher the score it receives.  (1.1, P.27)
>
> -- TOP's intended function is essentially like Google's ranking: it
> gives "how well" a given returned row matches a data set.  Hence, for
> a given dataset score would be different for different queries, and
> while a global quality measure might play a role, it certainly
> wouldn't be expected to dominate the response.  Or am I completely
> off here?
>
> And regarding Walter's interjection:
>
>> I thought that the difference between MAXREC and TOP is that MAXREC
>> requires an overflow indicator, while TOP would prohibit it.
>
> Interesting thought -- is it intended to work this way? [it doesn't
> in DaCHS, and a quick search in the 1.1 specs didn't give me anything
> pointing in that direction]
>
> Cheers,
>
>        Markus
>