Re: [UrbanSim-Users] location choice models: problem with sampling alternatives

Brian,

There are two kinds of jobs placed by the Distribute Unplaced Jobs Model:

1. Those that are not covered by the specification. For example in the
Eugene model, the specification for commercial ELCM contains submodels
(i.e. sectors) 1-8. Commercial jobs of the remaining sectors (i.e. 9-16)
are handled by the DUJ model. Analogous applies to the industrial jobs.

2. Jobs that were not placed via the ELCM due to capacity restrictions.
You should get an idea about the numbers by inspecting the log file (in
the ELCM section you should see something like 'Number of unplaced jobs:
xx (in 3 iterations)', one such line per submodel).

I suspect that the majority of jobs handled by the DUJ model belongs to
the first category. If it's a lot or not then depends on your
preferences, how the job placing should be done.

The answer to your last question is no. The developer model only builds
sqft if that gridcell was selected for building a project.

Hana

Brian Miles wrote:
> Hi Liming,
>
> I made a mistake. I wanted an indicator that would show me vacant
> commercial units by gridcell, especially those gridcells that had
> negative vacancies. To do this, I modified the function
> opus_core.misc::clip_to_zero_if_needed so that it would not clip to
> zero.
>
> After computing the indicator, but before running the Eugene model and
> our own model, I forgot to revert clip_to_zero to its correct behavior.
>
> Your answer was correct. I was the culprit. Sorry for the confusion.
>
> Now for a legitimate question. In our model, as well as in the Eugene
> model, we're seeing what seems to be a high number of unplaced jobs
> being placed by the Distribute Unplaced Jobs Model. Put more
> concretely, in both models, there are roughly 75,000 total jobs in the
> base year, and we're seeing on the order of 3,000 to 4,000 unplaced
> jobs at times throughout the simulation.
>
> My question is, is this "normal"? Is there an accepted threshold for
> a reasonable number of jobs that cannot be placed by ELCM?
>
> Also, in our model, we're seeing an accumulation of negative job space
> vacancy as time progresses. Does the developer model try to develop
> more job space, by building non-residential square footage, in areas
> that currently have negative job space?
>
> Thanks,
>
> Brian
>
>
> On May 2, 2008, at 3:23 PM, Liming Wang wrote:
>
>> Brian,
>> This means the array used as weight in the weighted sampling contains
>> negative (less than 0) values. In most case, the weight array is from
>> vacant units attribute, which could contain negative values if more
>> households are allocated to gridcells/buildings than their existing
>> units in the baseyear. In which model do you encounter these
>> warnings?
>>
>> I tried to run eugene sample project for both estimation and
>> simulation, and I didn't seem to get any of similar warnings.
>>
>> Liming.
>>
>> On Fri, May 2, 2008 at 11:07 AM, Brian Miles <Brian.Miles_at_uvm.edu>
>> wrote:
>>> Hello,
>>>
>>> After recently upgrading to the latest development version of OPUS, I
>>> noticed the following messages that appear when our location choice
>>> models run:
>>>
>>> WARNING: Problem with sampling alternatives.
>>> some values of rows [ 929 1797 2590 3657
>>> 3757 5265 7923 10166 12310] in weight_array are less than 0
>>>
>>> The locations differ across models, and sometimes there are hundreds
>>> reported in the message.
>>>
>>> I thought this was a strange message, so I decided to compare the
>>> output of our model to the Eugene example. I found similar "errors"
>>> in the Eugene model.
>>>
>>> My questions are: 1) what does this message mean; and 2) does it
>>> indicate a problem that needs to be corrected?
>>>
>>> Thanks,
>>>
>>> Brian
>>> _______________________________________________
>>> Users mailing list
>>> Users_at_urbansim.org
>>> http://www.urbansim.org/mailman/listinfo/users
>>>
>
> _______________________________________________
> Users mailing list
> Users_at_urbansim.org
> http://www.urbansim.org/mailman/listinfo/users
Received on Tue May 06 2008 - 14:39:52 PDT