MarkovReebs

MarkovReebs contains experiments for generating synthetic mobility trajectories from trajectory datasets using our Reeb graph-based generation methods and baseline methods.

The main workflow is:

1. Load Urban Anomalies or Geolife trajectories.
2. Build a multi-agent Reeb graph (MARG) from training trajectories.
3. Generate per-agent trajectories with SRG, HRG, or Markov-chain baselines.
4. Evaluate generated trajectories against held-out trajectories.

Setup

Install dependencies with uv:

uv sync

Download both of the necessary datasets. Set dataset locations in the environment or in .envrc:

export URBAN_ANOMALIES_PATH=/path/to/UrbanAnomalies
export GEOLIFE_PATH="/path/to/Geolife 1.3"

DATAPATH is also accepted as a fallback for Urban Anomalies. If these variables are not set, the loaders fall back to the historical defaults under /data/Datasets/.

Generate Outputs

Build a MARG pickle:

uv run python marg.py

Generate single-agent Reeb graph outputs:

uv run python srg.py

Generate hybrid Reeb graph outputs:

uv run python hrg.py

Evaluate

Run the evaluation script after generating the required files in outputs/:

uv run python eval.py

eval.py loads method outputs, optionally trims Geolife trajectories, computes agent and population metrics, and prints the result dictionary.

It also writes the captured results to:

outputs/table.log

Then it immediately runs the table formatter and writes:

outputs/generated_results.typ

Repository Layout

dataloader/ contains dataset adapters for Urban Anomalies and Geolife.

marg.py, srg.py, and hrg.py build multi-agent, single-agent, and hybrid Reeb graph outputs.

baseline.py and baseline/ implement the Markov-chain baseline.

evaluation/ and eval.py compute metrics and table-ready summaries.

visualize.py and visualizations/ contain exploratory plotting utilities.