Geographic Optimization with Bayesian Networks and BayesiaLab
Recorded on May 18, 2018.
- Presentation Slides (PDF, 13 MB)
- County-Level Commute Data (CSV, 138 KB)
- 95,000 Random Origin-Destination Pairs by ZIP Code (CSV, 7.7 MB)
- Hub Location Model (XBL, 586 KB)
With the recent release of version 7, BayesiaLab can now visualize the values of nodes in Bayesian networks on Google Maps. Beyond this convenient mapping capability, BayesiaLab offers several fundamental advantages in dealing with optimization problems in travel, transportation, and logistics. Instead of computing travel paths explicitly, we infer distances with a Bayesian network that was machine-learned from observed travel data, thus accelerating the search for an optimal business location, for example.
In this webinar, we present a complete modeling workflow from acquiring raw travel data to presenting the optimization results on Google Maps.
Optimizing travel routing has been a central topic in the field of Operations Research for many years. For a traveler, or rather his navigation system, this involves evaluating many possible paths between the origin and the destination and then selecting the shortest or perhaps the fastest route. To find an optimal location for a new retail store, however, we would need to evaluate many paths of many shoppers for a number of possible destinations with the objective of making it easily accessible. Clearly, that's a bigger computational task. Even more challenging is finding an optimal location for a transit hub, e.g., a freight distribution center, which requires evaluating many possible paths for many origin-destination pairs. It is easy to see that the explicit calculation of billions of routes can quickly become intractable.
We propose an alternative approach: Instead of calculating hypothetical route distances from map data, we machine-learn a Bayesian network from real-world travel data with BayesiaLab. Such a network approximates the joint distribution of trip-related attributes, including longitude/latitude of origin/destination and actual travel time/distance. Additionally, the Bayesian network automatically captures the frequency of the origin-destination pairs. As a result, we have a single model that compactly represents all road traffic. What is the advantage? We can now evaluate
|May 8–10, 2019||Singapore||Introductory Course (3 Days)|
|May 13–15, 2019||Sydney, Australia||Introductory Course (3 Days)|
|May 21–23, 2019||Paris, France||Advanced Course (3 Days, in French)|
|June 5, 2019||Washington, D.C.||BayesiaLab 101 Short Course (1 Day)|
|June 12–14, 2019||Seattle, WA||Introductory Course (3 Days)|
|June 17–19, 2019||Seattle, WA||Advanced Course (3 Days)|
Upcoming Seminars, Webinars, and Conferences
|Live Webinar||May 16, 2019||11:00 – 12:00 (CDT, UTC-5)||Human-Machine Teaming|
|Live Webinar||May 30, 2019||11:00 – 12:00 (CDT, UTC-5)||Causal Counterfactuals for Contribution Analysis — Explaining a Misunderstood Concept with Bayesian Networks|
|Live Webinar||June 13, 2019||11:00 – 12:00 (CDT, UTC-5)||Black Swans & Bayesian Networks — Jointly Representing Common and Rare Events|
|Please check out our archive of recordings of previous events.|