2026-05-21 Engineering Team

[joecastiglione]

Agenda

• sampling bias in accessibility
• random numbers

(Jeff might join late, don't wait)

[janzill]

Full EET PR, including Poisson sampling, at ActivitySim/activitysim#1064. As mentioned on Tuesday and in that PR, I believe disaggregate accessibilities are biased by log(sample_size) when using the sample correction factor in MTC_extended/SANDAG with MC or EET sampling, whereas it is not for Poisson sampling with the current implementation and the same spec. I will prepare some slides to go over this.

[dhensle]

New "faster" RNG is indeed faster, but still not quite as fast as the non-EET run. Below screenshot shows full sandag disagregate accessibilities model running single process and I think is a solid comparison for a model with a ton of EET draws.

Full profiling for a run with a 10k households running single process without sharrow on an RSG server shown below.
profiling_eet_legacy_10k_single_process.zip
profiling_no_eet_10k_single_process.zip
profiling_faster_10k_single_process.zip

runtime_comp_sandag_10k_sample_sp.xlsx