Estimation Enhancements

[dhensle]

Estimation work as part of ActivitySim's Phase 9B development effort.

• Implement multiprocessing for estimation mode
• Implement destination choice sampling in estimation mode
• Change the formatting and data written to EDBs
• Update Larch integration to accept new file formats
• Functionality to quickly test different specifications
• Improved larch reporting on estimated models
• Adding "predict" functionality with estimated models in larch
• Unit testing for the above features
• Updated documentation

[jpn--]

@asiripanich, this is a sharrow-related error. I'll take a look at it and see if I can figure out what's wrong...

Thanks @dhensle for your response.

I ran into an issue today while trying to add a new utility term to the tour mode choice spec of my tour mode choice EDB. This might be a bug or something that needs to be explained a bit more in the documentation. (However, I didn't have any issues modifying or adding new terms to the auto_ownership model. Despite this minor issue, I'm loving this new capability!)

When I tried to add a new line to my tour mode choice spec file, I kept getting this error:

modelname = "tour_mode_choice"

from activitysim.estimation.larch import component_model

model, data = component_model(
    modelname,
    edb_directory=f"output/estimation_data_bundle/{modelname}/",
    return_data=True,
)

> loading from output/estimation_data_bundle/tour_mode_choice/tour_mode_choice_coefficients.csv
> loading from output/estimation_data_bundle/tour_mode_choice/tour_mode_choice_coefficients_template.csv
> loading spec from output/estimation_data_bundle/tour_mode_choice/tour_mode_choice_SPEC.csv
> loading from output/estimation_data_bundle/tour_mode_choice/tour_mode_choice_values_combined.parquet
> unable to rewrite 'util_test' to itself

I tested this using example_estimation and 17_tour_mode_choice.ipynb from this PR.

I added these following lines:

tour_mode_choice_SPEC.csv

util_test,Drive alone not available for escort tours,1,coef_test,,,,,,,,,,,,,,,,,,,,

tour_mode_choice_coefficients_template.csv

coef_test,coef_test_eatout_escort_othdiscr_othmaint_shopping_social_work_atwork,coef_test_eatout_escort_othdiscr_othmaint_shopping_social_work_atwork,coef_test_eatout_escort_othdiscr_othmaint_shopping_social_work_atwork,coef_test_eatout_escort_othdiscr_othmaint_shopping_social_work_atwork,coef_test_school_univ,coef_test_eatout_escort_othdiscr_othmaint_shopping_social_work_atwork,coef_test_eatout_escort_othdiscr_othmaint_shopping_social_work_atwork,coef_test_school_univ,coef_test_eatout_escort_othdiscr_othmaint_shopping_social_work_atwork,coef_test_eatout_escort_othdiscr_othmaint_shopping_social_work_atwork

tour_mode_choice_coefficients.csv

coef_test_eatout_escort_othdiscr_othmaint_shopping_social_work_atwork,0,F
coef_test_school_univ,0,F

Any advice on what I might be doing wrong would be greatly appreciated. 😀

See full error

[jpn--]

@asiripanich, turns out you were doing nothing wrong, you just happened to find a bug -- a couple arguments were missing from the code to re-estimate the mode choice models. I've fixed it, and taken your edits to create an example / unit test of re-estimation on the tour mode choice notebook. If you try again it should work now. 😄

[asiripanich]

@asiripanich, turns out you were doing nothing wrong, you just happened to find a bug -- a couple arguments were missing from the code to re-estimate the mode choice models. I've fixed it, and taken your edits to create an example / unit test of re-estimation on the tour mode choice notebook. If you try again it should work now. 😄

@jpn-- Brilliant! I was away and haven't had a chance to test the new changes in this PR yet, but I'm looking forward to trying it out. I remember seeing this feature was planned for v1.4.0... is there an updated timeline for its release?

[bhargavasana]

Would it be possible to confirm that this issue still exists #595 ? Also #752 please.

[bhargavasana]

@dhensle please confirm that this PR fixes #897

[dhensle]

@dhensle please confirm that this PR fixes #897

Yes, closed that issue under the assumption this will be pulled in shortly.