Skip to content

MAST engine

MAST is built on the ABIT code base (msmobility/abit). Its own code is the package abm.models.maryland and the run class abm.RunAbitMarylandLos, in activity/engine/ABIT/src/main/java.

Classes

Class Role
abm.RunAbitMarylandLos entry point; args[0] = properties file; seed from -Dabit.seed
abm.models.ModelSetupMaryland wires the sub-models; mode.choice.model = tour_nl selects the tour mode model
abm.io.input.MarylandLosDataReader reads zones, skims, the land-use households, persons, jobs, dwellings, schools and workplaces
MarylandTashaGenerator generation and scheduling from the survey distributions; outings, return home, stop relocation, tours (tasha.generation=true)
MarylandDestinationChoice destination and stop logits over all zones, one log attraction term per model (dest.choice.model=mnl)
MarylandTourNestedModeChoice tour mode choice by segment; household stage (joint tours, car allocation)
WorkAttendance daily work attendance
MarylandFullModeChoice, MarylandTourModeChoice, MarylandFrequencyGenerator, MarylandStopModel, … earlier paths, kept but switched off

What happens per household

Households run in parallel, each with its own random stream.

flowchart TD
  A[Joint episodes<br/>frequency, start, duration, participants, place] --> B[Each member's agenda<br/>types in precedence order]
  B --> C[Outings from the workplace]
  C --> D[Clean-up of short work and lunch-at-home episodes]
  D --> E[Chains: return-home logit]
  E --> F[Main destinations and stops: location logits]
  F --> G[Tours and tour modes]
  G --> H[Household stage: joint tours together, cars allocated]
  H --> I[Outing modes; transit leg modes]
  I --> J[activities.csv, legs.csv]

Agenda order: WB, WAHB, PW, SW, RHW, SCH, JO_REC, JO_OTH, JS, IO_REC, IO_ACC, IO_OTH, IS. A conflict is resolved by shifting the episode (up to 30 min), then shortening it (to half), then redrawing it (up to 10 times).

Coefficient set: activity/parameters/

File Holds
run_params.csv model switches appended to the properties
tour_mode_choice.csv tour mode coefficients: work, school, shopping, other, joint shopping, joint other
parking_paid_share.csv paid-parking share by destination area
dest_choice_mnl.csv destination and stop logits (primary, stops, school/daycare, work business, home workers, outings)
tasha/tasha_frequency.csv, tasha_start.csv, tasha_duration.csv survey distributions
tasha/tasha_joint_*.csv joint episodes: participants and propensity
tasha/tasha_return_home.csv return-home logit
tasha/tasha_subtour.csv outing generation and mode
tasha/tasha_transit_legs.csv transit leg modes
tasha/tasha_params.csv scheduler parameters, worker-type shares, file names
TERM_MAP.csv every estimated term → engine file and variable

activity/build_parameters.py writes this set from the estimation outputs and records a checksum of every source in build_state.json.

Run

python activity/run_activity_model.py myrun --coef activity/parameters --seed 1 --sample 0.1 --threads 6 --xmx 13g --year 2024

This copies the coefficient set into the run folder and starts java -Dabit.seed=N -Djava.library.path=… abm.RunAbitMarylandLos <properties>.