Page 1 of 10

[NEW] Best Continuation algorithm has been changed

PostPosted: Sun Sep 14, 2014 9:41 am
by PesachZ
It has recently come to our attention that in the past week or two Waze has altered the algorithm used to determine the Best Continuation at a fork in the road. This was first posted about in the wiki update thread, but I'll leave that thread for discussing the upcoming necessary changes to the wiki to reflect this change. I'm creating this thread to alert other editors to this fact, which is already causing lots of URs.

In short: the algorithm looked for a match from the segment preceding the junction, to either of the segments after the junction. It first checked for a matching primary name, if it didn't find any it then checked for matching segment types. If a match was found that was the Best Continuation, otherwise both segments would get an instruction.

The good news:
The Best Continuation is only a factor when 2 or more segments leave a junction at less than 45° turn angles. (e.g. a fork in the road, an exit from a highway, a Wayfinder, etc.) This has no effect on turns which would give a "turn right/left" instruction, only ones which would give a "keep/stay/exit right/left" instruction. For a clearer understanding of the whole Best Continuation topic, please read the wiki page linked above.

The other news:
Alternate names now play a role in factoring which segment will be the Best Continuation! This is a very recent development, and can cause many highway exits, wayfinders, and road splits which were previously giving the proper instructions to have suddenly either stopped giving an instruction to a segment which needs one, started giving an instruction to a segment which is the clear continuation and doesn't need one. We still do not fully understand the full extent of this change, and how different variables are being taken into consideration in the algorithm. There is extensive testing underway (which will be documented in the wiki update thread) to determine all these variables.
What we have tested thus far, the following matches will be BC in order of preference:
  • Two primary names AND their road types
  • Two alt names AND their road types
  • Two primary names only
  • Two alt names only
  • When there is any alt name on the preceding segment, there can be a cross-match between an alt name and a primary name (the alt of one segment matches the primary of the other)
    • If one side is a primary match, and the other is an alt match and a road type match, the side with alt and type match is BC (This is the first time we have seen a primary name match not be the BC)
    • When there is no alt name on both segments in a pair, alt names are not considered for matching
    • City names do not seem to affect this at all
    • There is no preference given to the order of the alt names
EDITED list for clarity

What do you need to know for now.
If you see URs, complaining of this behavior, please investigate them carefully, run the route through Livemap to be sure if the instruction has now changed, even if all the segment names appear correct in WME. Bear in mind that any Continue instruction shown in Livemap with a straight arrow is NOT given in the client. If you find a problem caused by this change, please don't edit anything, and post the PL to this thread so we can investigate it to help us learn the full extent of this change. After reviewing it, we will fix it to make it work properly again.

Re: [NEW] Best Continuation algorithm has been changed

PostPosted: Sun Sep 14, 2014 11:08 am
by Shaq23
Just came across a UR in Maryland, where the driver did not get an an instruction to keep right for the I-95 N. Livemap did not give an instruction as well. Looking at the S-In and S-out, they both have a matching alt name of "I-95 N".

Side-note: This Wayfinder seems like it should be set as a Right FWY Split, and the stubs should be Ramp type.

Re: [NEW] Best Continuation algorithm has been changed

PostPosted: Sun Sep 14, 2014 11:42 am
by PesachZ
Shaq23 wrote:Just came across a UR in Maryland, where the driver did not get an an instruction to keep right for the I-95 N. Livemap did not give an instruction as well. Looking at the S-In and S-out, they both have a matching alt name of "I-95 N".

Side-note: This Wayfinder seems like it should be set as a Right FWY Split, and the stubs should be Ramp type.

Fixed, thanks

Re: [NEW] Best Continuation algorithm has been changed

PostPosted: Sun Sep 14, 2014 4:39 pm
by asterix06
Why anyone of Waze Staff didn't asked anything to us , before to make such a type of important changes on the alghoritm ?

Many many users have been faced with strange routing behavior , where we could hardly give an answer and creating a lot of URs as well.

Re: [NEW] Best Continuation algorithm has been changed

PostPosted: Sun Sep 14, 2014 5:34 pm
by PesachZ
asterix06 wrote:Why anyone of Waze Staff didn't asked anything to us , before to make such a type of important changes on the alghoritm ?

Many many users have been faced with strange routing behavior , where we could hardly give an answer and creating a lot of URs as well.

That's the million dollar question, or at least after so we know the specifics of to work around the change. But, alas we do what we always do, and try to figure it out for the best. Maybe this time they'll chime in and help us out.

Re: [NEW] Best Continuation algorithm has been changed

PostPosted: Sun Sep 14, 2014 11:51 pm
by PesachZ
I edited the list in the OP and added a bit to clarify what we know so far.

Sent using Tapatalk for Android 4.4.2

Re: [NEW] Best Continuation algorithm has been changed

PostPosted: Mon Sep 15, 2014 12:08 pm
by jaywazin
I suspect that since the best continuation plays a big point in Wayfinders, the Wayfinder section of wiki may no longer be valid.

I've been trying to get a couple wayfinders in my region to work for about a week or two and was about to post a request for help. Do you think this is a related - and if so should the Wayfinder section of the wiki be edited to reflect the fact that this is again being reformulation?

Re: [NEW] Best Continuation algorithm has been changed

PostPosted: Mon Sep 15, 2014 12:29 pm
by milkyway35
Please note : something valid for USA/Canada is not necessarily valid for World or Israel.

These 3 platforms can have different behaviours.

[NEW] Best Continuation algorithm has been changed

PostPosted: Mon Sep 15, 2014 12:40 pm
by asterix06
Marc is correct !
Are we sure modifications and algorithms are the same for all platforms ?


Inviato dal mio iPhone utilizzando Tapatalk

Re: [NEW] Best Continuation algorithm has been changed

PostPosted: Mon Sep 15, 2014 2:53 pm
by PesachZ
jaywazin wrote:I suspect that since the best continuation plays a big point in Wayfinders, the Wayfinder section of wiki may no longer be valid.

I've been trying to get a couple wayfinders in my region to work for about a week or two and was about to post a request for help. Do you think this is a related - and if so should the Wayfinder section of the wiki be edited to reflect the fact that this is again being reformulation?

Yes it can affect certain Wayfinders, and that section will be modified shortly to reflect this, once we know all the details we need. For the most part that section will remain the way it is, with the exception of some added details where alt names are concerned.

If you could be so kind, please post the PermaLinks here to the Wayfinders you are trying to get to work, as the time frame matches perfectly. We would love to take a look at them and see if we can learn anything new from them. We can then also help you set them correctly so they are not affected by this change.

milkyway35 wrote:Please note : something valid for USA/Canada is not necessarily valid for World or Israel.

These 3 platforms can have different behaviours.

Up until now this portion of the code, and its associated behaviors were common to all servers, we have no reason to believe that that has changed, though anything is possible. If you want to setup a test in your country to confirm I'd happy to help you through it. Just send me a PM.