accessibility task

[bstabler]

I want to make sure that we start with the right version of MTC's TM1 accessibility script. Please confirm that we need to re-implement this script into ActivitySim as a new fully featured sub-model.

https://github.com/MetropolitanTransportationCommission/travel-model-one/blob/master/model-files/scripts/skims/Accessibility.job

[DavidOry]

I read the scope as translating this script as part 1 and computing destination choice logsums as part 2. I am personally far more interested in part 2 than part 1.

[bstabler]

Yes, the first part is to translate this script, thanks for confirming. I think it makes sense to do the second part after (or at the same time as) task 8. Ok?

[DavidOry]

I agree.

[toliwaga]

I have the translated script implemented and am getting very close but not exactly duplicate, results in python as the accessibility table supposedly created by mtc_accessibility.job.

I did notice one thing in the original job script:

In mtc_accessibility.job at line 191 we have:

     ; compute the decay function for peak transit accessibility if a round trip path is available (zero otherwise)
     if(trPkTime_od > 0 && trPkTime_do > 0)

        trPkRetail = retailEmp * exp(_kTran * trPkTime)
        trPkTotal  = totalEmp  * exp(_kTran * trPkTime)

     endif

But in mtc_accessibility.job at line 210 we have:

     ;   - off-peak, round-trip time
     trOpTime = trOpTime_od + trOpTime_do

     ; compute the decay function for off-peak transit accessibility if a round trip path is available (zero otherwise)
     if(trOpTime>0)

        trOpRetail = retailEmp * exp(_kTran * trOpTime)
        trOpTotal  = totalEmp  * exp(_kTran * trOpTime)

     endif

I am thinking we should have instead:

if(trOpTime_od > 0 && trOpTime_do > 0)

Line 214 assumes that it will always be possible to travel in both directions between any two skims at midday.

This is not the case. There are 28006 OD taz pairs for which one of these is 0 and the other is not.

orig=6, dest=362, for instance:

INFO - activitysim.defaults.models.accessibility - trOpTime_od[6, 362] = 224.6554
INFO - activitysim.defaults.models.accessibility - trOpTime_do[6, 362] = 0.0

[bstabler]

this looks like a bug to me so let's fix it in our version

[toliwaga]

Accessibility method now working with UEC-style expression file. I am getting pretty similar results to the accessibilities that were calculated by the job, and exactly the same results of my previous numpy matrix implementation.

The fact that I get exactly the same results as the job version for non motorized which makes me think maybe we have slightly different versions of the transit and traffic skims…

Now that this is working, we need to decide a few things:

• do we overwrite the existing accessibilities table (with it’s nasty all-caps column names lik AUTOPEAKRETAIL) or do we go back and change all the expressions files to use updated column names?
• Do we accept the slightly different accessibilities or investigate why ours differ slightly?
• If we accept the new values, I guess we go back and simply replace all the regression tests with updated values based on the new accessibilities..
• Right now the accessibilities expressions only support python (and not pandas) evals. Not much reason to support pandas (since we want to access skims) but is this ok?
• Currently calling skim access object skim and skim_t and data frame `df``, but this could easily change.

[toliwaga]

I am submitting a pull request for the current accessibility branch, which I believe fulfills aprt one of this task: it implements the translation of MTC's TM1 accessibility script as expression files that can easily be extended or modified without updating python code. This is implemented as a sub-model that is run before other models.

Currently it does not override the existing accessibility table values, so results can be compared to the existing values. Eventually we will presumably want to do so and update the regression tests accordingly.

[bstabler]

• do we overwrite the existing accessibilities table (with it’s nasty all-caps column names like AUTOPEAKRETAIL) or do we go back and change all the expressions files to use updated column names?

go back and update the expression files

• Do we accept the slightly different accessibilities or investigate why ours differ slightly?

accept

• If we accept the new values, I guess we go back and simply replace all the regression tests with updated values based on the new accessibilities..

yes, update all the regression tests

• Right now the accessibilities expressions only support python (and not pandas) evals. Not much reason to support pandas (since we want to access skims) but is this ok?

yes

• Currently calling skim access object skim and skim_t and data frame df`, but this could easily change.

change to skim_od and skim_do. And change all expression files to use something smarter like skim_od and skim_do as well. See #105

• Need to fix accessibilities.csv so it has all the commas for all columns for the comment rows as well

we need to do this as well

[toliwaga]

The six bullet items in the comment above are done and pushed to branch. I will pull them to master branch and close this issue - we should create a separate issue for logsum choice (part 2) as we won't be doing it until Task 8.

[bstabler]

While writing the accessibility module documentation, I noticed a few things to revise:

1. We should trace the temporary expressions as well, such as _auPkTime
2. We don't need to trace accessibility.full.csv since this goes into the data store and we don't care about the differences w.r.t. the old/original accessibilities.
3. I don't understand the accessibility.csv trace file since it has two columns - one for the origin zone and one for the destination zone? I was only expecting to see accessibility.result.csv, which has one column - one for the OD pair. I think we can drop this output.

[toliwaga]

I fixed the trace files in nested-logit branch.

Will trace temp expressions soon...

[toliwaga]

Added tracing of temps in pull request #109