Skip to content

Comment on Optimization Solver as a Serviceparent

Comments

I beg to differ. Timefold is used daily in real-world operations.

Neither our clusters in the US or EU can afford to go down for a minute, or business operations in logistics, retail and healthcare are impacted.

They do not adequately account for traffic, driver preferences, or other factors that require a continuous feedback

- Traffic: supported

- Driver preferences: the APIs support

-- Area affinity (soft) and geo fencing (hard)

-- Fairness and load balancing of work (soft)

-- Overtime (soft and hard)

-- Seniors not doing boring junior tasks (soft)

- Continuous feedback: the APIs support

-- Real-time rescheduling (warm starts) as actual execution data comes in.

-- Recommend assignment: scheduling in survival mode

-- Pinning: user stays in control through overriding assignments

-- Explainability: why a certain decision was made

-- Insights: what are the bottlenecks in my schedule - what type of employees should I hire or upskill.

That being said: I fully agree that the solver industry as a whole has focused for far too long on global optima for academic requirements, instead of real-world use for the actual business requirements, and how to deal with business objective changes each quarter.

AboutSource Built by g1lg1l

Hackerly is an independent reader for Hacker News, built on the public HN API. Not affiliated with Y Combinator.