Sort by
Author
Post time
Subject
Display
All posts
1 day
2 weeks
1 month
3 months
6 months
1 year
Post Reply

Roads auto-fixing (geometry, directionality, ...)

Post by freddiechopin
Somewhere (wiki? FAQ? doesn't matter now) it's said, that Waze will fix the roads based on other users (statistically). The geometry thing is pretty cool, but the directionality thing is pretty screwed up right now. I've fixed some roads just 2 days ago in the Cartouche and today I've found that some segmnts were converted to two-way (from one-way), some had the directionality reversed... (two examples marked in this view) I think it's safe to assume that this changes are an effect of ONE merging process. Now - here in Poland we have so little Wazers, that I can bet that a totally insignificant number of people used that road on this day (5?). Now, my fix two days ago was to align the roads with GPS tracks so they are pretty accurate, so how many of these small number of people would have left WRONG tracks? Now, the directionality was changed only on ONE direction, so this very-small number of people has to be divided by two (let's assume that an equal number of people travel in both directions). These were not my tracks, as I drove in the "right" direction. To be honest I'm not sure anyone drove that road on this period... "Somewhere" it was said that Waze tries to be conservative about such changes, and numbers like 100 users are given.

So what's the problem here? If you check countries with developed Waze maps (let that be Slovakia - example [segments of freeway]) you notice that almost ALL roads are locked. This is to prevent Waze's algorithms from screwing people's work. Now these features make perfect sense and the idea behind them is good, but in reality - as I've tried to prove - they don't help at all, and people have to take additional measures to prevent them from working. You can even check the discussions on the forum where people are advising other users to tick "Locked" after EVERY edit.

Waze team - either disable those algorithms at all, or fix them (for example by really adding some HIGH threshold, because obviously now there is none...).

I hope that developers or support are reading this forum, if someone thinks otherwise tell me - I'll send a link to this thread via e-mail.

4\/3!!
freddiechopin
Posts: 256

POSTER_ID:1340685

1

Send a message

Re: Roads auto-fixing (geometry, directionality, ...)

Post by dmcconachie
We can't really see deep enough into the data to make a judgement, I'm afraid it's one for alpha@waze.com again!
dmcconachie
Posts: 6119
Has thanked: 55 times
Been thanked: 181 times
Send a message
dmcconachie • Former Global Champ

UK Wiki • UK FB Page • Current status

Re: Roads auto-fixing (geometry, directionality, ...)

Post by dmcconachie
freddiechopin wrote:Well, there's also a solution of fixing the algorithm and it's the best solution.

4\/3!!
In your opinion!
dmcconachie
Posts: 6119
Has thanked: 55 times
Been thanked: 181 times
Send a message
dmcconachie • Former Global Champ

UK Wiki • UK FB Page • Current status

Re: Roads auto-fixing (geometry, directionality, ...)

Post by dmcconachie
I agree Alan, hopefully just temporary though!

In my most humble opinion the ideal solution would be that editing auto-locks directionality but not geometry but both have individual checkboxes on the edit panel!
dmcconachie
Posts: 6119
Has thanked: 55 times
Been thanked: 181 times
Send a message
dmcconachie • Former Global Champ

UK Wiki • UK FB Page • Current status

Re: Roads auto-fixing (geometry, directionality, ...)

Post by dmcconachie
Man, you complain too easy! ;-)

Here's one which was driving me nuts, basically a very large junction with 5 intersecting roads. It used to be a roundabout but was recently converted to signal controlled. Nothing looked right and I tried many iterations so I set about creating a monster...

https://world.waze.com/cartouche/?zoom= ... FTFFFFTTTT

It routes perfectly as far as I can tell and on the client map appears as quite a large block rather than a heap of organised spaghetti.

So yes I would vote for your junction area rather than single junction point!
dmcconachie
Posts: 6119
Has thanked: 55 times
Been thanked: 181 times
Send a message
dmcconachie • Former Global Champ

UK Wiki • UK FB Page • Current status

Re: Roads auto-fixing (geometry, directionality, ...)

Post by freddiechopin
Well, sure - that's some type of solution. But the REAL solution would be to improve the auto-fixing algorithms to behave as they should, because - I'll say it again - the idea behind them is very very good, but the implementation has some major flaws, in my opinion too low thresholds or wrong threshold calculations. What's the point of keeping this mechanism (and running it, probably pretty heavy code) if everyone would just tick "Locked" everywhere to prevent it from working? For example I don't mind geometry changes (makes the curves very smooth), but the directionality change should require a LOT of input validating such change - for example a hundred tracks that are more or less equally divided in two directions (so ~50 going in one direction and ~50 in the other) and that REALLY match that street segment, because we know that GPSes in our phones are not perfectly accurate.

Here that may be as little as me traveling one way and another (just ONE) wazer travelling the other way - this surely is not enough to validate such big change as directionality, taking into account that the connectivity (turn restrictions) and geometry (split lanes for major road) suggest such change to be wrong. This algorithm has to be improved so that locking segments would not be necessary - it would even be WRONG to lock them.

4\/3!!
freddiechopin
Posts: 256
Send a message

Re: Roads auto-fixing (geometry, directionality, ...)

Post by freddiechopin
I've sent a link to this topic to Waze support via e-mail.

4\/3!!
freddiechopin
Posts: 256
Send a message

Re: Roads auto-fixing (geometry, directionality, ...)

Post by freddiechopin
unwallflower wrote:Like I said... the Waze auto-correcting feature is only as good as the data it has to work with. If the GPS tracks it receives say that people are driving in both directions on the same segment of road... well, that's what it has to go on.
I see it completely the other way around. The data is as it is - sometimes good, sometimes bad. If Waze's auto-fixing algorithm are going to be enabled, than these algorithms must work well on such data as users are able to provide. You say that data is not reliable and that it's not the algorithms fault. Yes - data is not perfect, but it won't be any better, so this leaves the algorithms to be tuned up.

As I said - changing direction because few (!) tracks match is not good and that should not be excused, because that threshold is way too low, and everyone of us can provide examples. Take a look at what I posted in the opening message - http://world.waze.com/cartouche/?zoom=4 ... =103380551
Zoom in, enable GPS, color arrows by azimuth and count the arrows that are in the wrong direction - 5. And like 1000x more arrows in the right direction. And 1000x more yellow arrows miraculously shifted a few meters east. I say that these marked segments are too much to the west! The algorithm is obviously wrong to change the directionality!

Still I think that tracks should not be locked, the algorithm should be fixed.

4\/3!!
freddiechopin
Posts: 256
Send a message

Re: Roads auto-fixing (geometry, directionality, ...)

Post by freddiechopin
The same is true for situation described by rl_pl on previous page:
http://world.waze.com/cartouche/?zoom=6 ... FFFFFFTTTT
enable GPS points and there are NO wrong-way arrows on affected segment. So how does that happen?

4\/3!!
freddiechopin
Posts: 256
Send a message

Re: Roads auto-fixing (geometry, directionality, ...)

Post by freddiechopin
rl_pl said that this happened several times. So statistically the tracks you see represent the average and most likely in a year the image would be more or less the same - with (say) 60% tracks in line, 30% (say) 5m off and 10% more than (say) 5m off. Yet there are NO wrong-way tracks. The black arrows are not even close to the segment.

4\/3!!
freddiechopin
Posts: 256
Send a message
Post Reply