Re: [UrbanSim-Users] Household location choice estimation

Brian,

brian voigt wrote:
> Thanks Hana, that does clear up some confusion. I have a couple
> follow-up questions. First, I still don't understand why the
> households_for_estimation and households tables are internally attached.
> Are the tables being joined or appended? I see where you distinguish
> between the agent_set and agents_for_estimation_table_name in
> urbansim/configurations/household_location_choice_model_configuration_creator.py,
> but am still puzzled about why the tables need to be attached.

If the argument 'join_agents_for_estimation_with_all_agents' (in the
configuration constructor) is True, the table households_for_estimation
is (internally) appended to the table households. We do this in order to
have the regional demographics available to the estimation procedure.
For example, if you use a variable
'percent_minority_within_walking_distance', you don't want to compute it
using only the subset data, right? Ideally, we would keep two tables -
one for computing such variables, one for the estimation. But that is
difficult to implement in the Opus framework (also because one would
have to distinguish what variable to compute on which of those
datasets). Our reasoning is that the number of households for estimation
will usually be much smaller that the whole population of households and
thus, appending them would not destroy the regional characteristics.

> Also, if
> I include the attribute INCOME (or AGE_OF_HEAD, PERSONS, etc) as
> urbansim.household.income in the HLCM specification, is that referring
> only to the data from the households_for_estimation table?

Yes, the estimation procedure extracts values of the attribute that
belong to households_for_estimation.

> Finally,
> although you suggested making changes directly to
> household_location_choice_model_configuration_creator.py, I'm wondering
> if there is a way that I can pass these arguments in a configuration
> file (so that I don't mess up the source code).

Unfortunately, this is still a little bit difficult. I'm in middle of
re-writing the estimation support so that such things will be possible,
and that one can configure and launch an estimation from a GUI. Right
now, one can make changes either in the creator file or in
urbansim/configs/general_configuration.py where the creator class is called.

Hana

>
> Hana Sevcikova wrote:
>> Brian,
>>
>>
>>> Perhaps I am confused about the function of the estimation procedure. I
>>> understand that the number of movers and the unplaced agents are derived
>>> by the Household Relocation Model, but how do these movers and unplaced
>>> agents relate to the records in the households_for_estimation table?
>>> More specifically, how can there be more movers / unplaced agents than
>>> there are records in the households_for_estimation table?
>>>
>>
>> In the default estimation behavior we try to create the same situation
>> as in the simulation of HLCM which runs in 12 chunks (corresponding to
>> 12 months). So we let the HRM choose movers (from the whole household
>> set) and unplace 1/12 of them (randomly chosen), in order to have
>> similar vacancy as in the simulation. Households in
>> households_for_estimation are unplaced too, but after the their location
>> is captured and passed to the estimation procedure. Note that the
>> households_for_estimation table is by default internally attached to the
>> households table, in order to have information about the whole population.
>>
>> As I said, this behavior is the default one and can be changed. You can
>> look at
>>
>> urbansim/configurations/household_location_choice_model_configuration_creator.py
>>
>> I just added comments to the arguments (so please get the newest file
>> from svn). You can then experiment with changing the default values here.
>>
>>
>>> Perhaps too I
>>> don't understand correctly what the dependent variable is that we are
>>> modeling. I thought the estimation procedure evaluated the location
>>> choice for each household in the household_for_estimation table compared
>>> to the attributes of the other records in the choice set. Is this correct?
>>>
>>
>> It is correct.
>>
>> I hope this clarifies it a little bit.
>>
>> Hana
>> _______________________________________________
>> Users mailing list
>> Users_at_urbansim.org
>> http://www.urbansim.org/mailman/listinfo/users
>>
Received on Wed Mar 05 2008 - 15:16:52 PST