AgentAccess

**Paper out now in CEUS: https://doi.org/10.1016/j.compenvurbsys.2026.102543

Authors: Alexander Michels

This project is working towards understanding accessibility through Agent-Based Modeling (ABM).

---

Quick Start

Run setup.sh to unpack the data and setup the environment.

Functionalities:

Cost Functions

Our cost functions take in the following arguments:

• time (t) the time from location i to location j
• utilization (u) the demand/supply ratio at location j

and there are a few parameters available for our cost functions:

• beta ($\beta$) for gravity - exponent for time
• tau ($\tau$) for raam - normalizes time

some parameters are calculated by on input data including:

• rho ($\rho$) for ramm - global inverse demand/supply ratio, normalizes supply/demand ratios

Agent seek to minimize the "cost" (in travel/waiting time). The functions we hare implemented are:

• gravity given by $$t^{\beta} * u$$
• raam given by $$\frac{u}{\rho} + \frac{t}{\tau}$$
• time given by $$t^{\beta}$$

Multiple Travel Modalities

Agents in the framework can have different travel modalities/travel times. This is achieved by setting cost_df to a DataFrame with multiple travel-cost columns and then passing a list of columns into cost_field.

• You can set the proportion of agents with each travel modality using the cost_prop to pass in a set of weights (summing to one) for each field (where cost_prop[i] is the weight for cost_field[i]).
• You can have the proportion vary by geospatial unit by passing in cost_prop_fields and we will assign agents proportionally to these fields for each spatial unit.

Parameters

The model has a few parameters you can play with:

• memory: the number of time steps agents will "remember" when determining which place they should utilize. If memory == 0, they will "remember" everything.
• pop_per_agent: we will give each spatial unit population / pop_per_agent agents rounding up.
• top_k: agents will choose between the top_k best options, proportionally to their cost. So if k=1, they always choose the minimum cost option. If k=3, they choose between the three best options with probability inversely proportional to their costs. For example, if we have locations [i,j,k] with costs [1,2,2] we first invert the costs to get [1, 1/2, 1/2], then normalize to approxiately obtain [0.5, 0.25, 0.25] which are the probabilities the agent chooses [i,j,k] respectively.

---

Todo:

• Fix code to allow for incomplete travel dicts (i.e. for each GEOID get the possible destinations and only consider resources at those destinations)
• Update statistics to better handle agents who were unable to make a choice