Add more place classification

[Joxit]

Background

When I tried a acceptance test on the API with the pelias_parser branch, I found some regressions. One of them was University of Hawaii at Hilo. I found that this was not classified as place, so I tried to added it.

In my first try, I added the correct classification in classifier/scheme/place.js but nothing happend. So I investigate a bit more and I found an exception on span containing a StopWordClassification in the WhosOnFirstClassifier.

What to do next ?

In order to limit the side effects, I preferred to add a penalty instead of completely removing this exception. With this, no test fails.

The parsing of University of Hawaii at Hilo is not at 100% correct, but it's better :)

If I completely remove the exception, 2 tests are failing, Rue de Paris and Avenue de Sainte Rose de Lima which should be streets.

I put this in standby at the moment in case it is necessary to keep the exception.
I am open to discussion / review 😄

[missinglink]

hmmm yea, I would have preferred not to have that logic in the WhosOnFirstClassifier.
I think it makes the code harder to debug and reason about why it is behaving the way it is.

I'm also not sure how I feel about confidence = confidence / 2, I was hoping that we could avoid modifying existing tokens and existing classifications but maybe that's 'wishful thinking'.

I'd like to try and refactor some of this 'smart' code to be clearer and more predictable, and try to be less clever if there is a more transparent way of doing the same thing.

I also had a look at the University of Hawaii at Hilo test case and gave up ;)
I don't think we should merge the test case in this PR because it's incorrect, I'd prefer to leave it commented out than commit incorrect tests.

So yea 🤷‍♂ I don't know how to get that venue to parse correctly yet, maybe we just think about it a bit more and a simpler solution will present itself?

[Joxit]

confidence = confidence / 2 is totally random, I thought maybe we could decrease confidence progressively. It's bit like what is done in Street*Classifier, but instead of static values we have dynamic ones (which can be in fact replaced by 0.5).

Okay, I commented the test in the hope that it will be solved one day 🤞.

The addressParts.map piece of code is awesome ! 😆

[missinglink]

👍