[Photon] Next steps

[Yohan Boniface]

On 06/17/2014 07:00 PM, Christoph Lingg wrote:
> I made a quick try, i did not get fuzzy to work.
>
>> >But then cross_fields is not the silver bullet, because you weigth*per field*  (which is nice), but you also want to weight differently if the query was fuzzy or not, which you can't do with a multimatch
> in my perception, cross_field is just a more elegant way (field specific boost) and more efficient way (no redundancy) of the collector. If fuzzy worked for instance I guess it would perfectly fit for the should branch herehttps://github.com/komoot/photon/blob/positivescoring/website/photon/app.py#L40
>

It is elegant in theory, but in practice you don't have enough control imho.
The tricky part at this point in the work is to control fuzzy, because 
you don't want a fuzzied result to have more weight than a non fuzzy one 
(unless other boost makes a real difference), and tf/idf: having twice 
time Paris in a row doesn't make it more relevant than "Paris" itself 
when you search for "Paris", and you don't want "Parvis" to overboost 
everything just because it's a rare term.

> now you only boost raw hits on name and the rest, this could be done more specifically.
>

I think that's just a matter of finishing, not the actual challenge.

>> >If you want to work on search logic, you should work on the positive scoring branch.
> that’s why i asked you in the first mail if you had some time tomorrow (maybe 15 min) to get coordinated, also thinking about how to contribute on thursday/friday.

Sure! Not the best day for me but let's try!
What works best for you? Should be easier for me around 11 or after 22 
roughly, but let's see :)