Bunch and Gap

[Theo Armour]

Hello Urdacha Contingent

The following could be a draft of what could be sent to the Meet-Up List and Google Group.

Any thoughts?

Theo

***

On 23 February, at the Urban Data Challenge Hackathon in San Francisco, the Urdacha team had the pleasure of sitting down with Chris Pangilinan of the SFMTA.

During the course of our chat, Chris was clear and straightforward about one of the things he would be delighted to have available. Chris said he would like to see the bunches and gaps between Muni vehicles on their routes.

We have taken this to mean two things:

1. There is a visual replay of the data
   
2. The bunches and gaps are indicated or highlighted during the replay

The Urdacha team has taken this wish list item as its command line and has been working towards achieving this goal ever since. Moreover the team has dedicated its work to producing this visualization equally across the datasets of the three cities. 

If you have been following the progress the the Urdacha team, you know that replaying the CSV data files in user-manipulable 3D is well underway. Yesterday's update is visible here:

http://jaanga.github.com/urdacha/haxis/r4/

With this app, you can visualize the data from the three cities replay with a wide variety of parameters and from any angle or perspective.

As fun as it may be, though, hAxis does not point out or highlight any gaps or bunches that may be occurring. Of course, humans can identify bunches and gaps during the replay, but this is not all as the same as issues being recognized and highlighted by algorithm.

Well, as of this afternoon, Urdachas has algorithmic bunch and gap capability.

Kindly have a look at:

http://jaanga.github.com/urdacha/haxis-add-ons/bunch-and-gap/

It's a bit ugly and slow, nevertheless this app identifies bunches and gaps between vehicles in all three cities.

It's also quite amusing. Both Geneva and Zurich have very high frequency services in the central business districts. The app pinpoints gaps and bunches even in services where there's a vehicle every minute.

How can we do this? The algorithm is quite straightforward. It creates a moving average of the five most recent vehicle stop times at all the stops. Any vehicle that arrives fifty percent faster or slower than the average is flagged. If, however, a vehicle takes several time longer than normal it is ignored. Perhaps the was a breakdown or a schedule change.

In any case, it's all early days and just a first foot forward. There are a number of issues that could be resolved and many features that could be added. And, frankly, we don't yet have concrete evidence that our algorithm is valid.

But if Chris, or you, want to click into the algorithm, it's all free and open-source and available here:

https://github.com/jaanga/urdacha

Enjoy!

Theo

 

[Cynthia Armour]

Hi, 

I think we could be less direct about making this specifically for a request made by Chris. I would say that the idea of helping identify gaps and bunches was appealing to Chris and other attendees of the hackathon. 

In terms of understanding the app - I can see where we have the total number of bunches and gaps per route - is the details regarding each bunch/gap on the left? How can we help people read this information? 

Cynthia

--
You received this message because you are subscribed to the Google Groups "Urdacha" group.
To post to this group, send email to urd...@googlegroups.com.
Visit this group at http://groups.google.com/group/urdacha?hl=en.
 
 

[Theo Armour]

Cynthia

Thanls fr the quick reply

>> I think we could be less direct about making this specifically for a request made by Chris. 

>>I would say that the idea of helping identify gaps and bunches was appealing to Chris and other attendees of the hackathon.

I am not exactly sure how I would edit the words to convey this.

The quickest way to get things done is to suggest the actual words you would use/do the edit you would do.

Also

Chris is a judge not just an attendee

I did not hear anybody at the hackathon or the original get-together talk about gaps and bunches, 

>> s the details regarding each bunch/gap on the left? How can we help people read this information? 

Given the parameters I chose, Zurich has over 40,000 bunches and the data is virtually indecipherable to a normal human - things like many rows of 2,000 numbers of six characters each. The proof is the source code.

And in the visualization I will try to build tomorrow...

Love,

Daddy