3 ms·

> Algos to solve those both exact methods and heuristics are super useful in real world applications, and very underappreciated by the HN crowd IMO. There just

> Algos to solve those both exact methods and heuristics are super useful in real world applications, and very underappreciated by the HN crowd IMO. There just aren't many situations where they're useful. Time scheduling, logistics, operations, there are only a handful of places where these problems are so important that they need to be solved optimally or nearly optimally.

cjalmeida 3y ago
IMO, the main benefit of those exact algos are the they provide optimality bounds, allowing you to make informed decision to continue or stop and provide a non-optimal solution. This is valuable in many practical situations. >Time scheduling, logistics, operations... Add marketing (pricing, assortment, ...) and finance (portfolio, captial budgeting, ...) to those. And those are huge domains, moving trillions of USD globally each year. Many companies (think most of the Global 2000) benefit from a couple % improvement on their operations that lead to many $$ in savings.
- eru 3y ago
  Even just arranging UI elements around a page could be done with these solvers, instead of coding up heuristics to satisfy your constraints by hand.
kj4211cash 3y ago
I don't think this is true. It's just that there aren't many optimization practitioners and the OR community has done a terrible job promoting itself. The value in scheduling and logistics is well established. So optimizers tend to congregate. But optimization could be applied to lots of other domains successfully. ML is just minimizing a loss function. I see simulation and post-hoc analyses being applied in numerous situations where optimization would be more efficient and produce a guaranteed optimal result. RL on simple, static problems is really an MDP but I've seen organizations fail to realize this. Etc.
- Karrot_Kream 3y ago
  How often do these come up? I studied some OR stuff in grad school when I was evaluating solvers and brushed up against the OR tools community for some research ideas and from what I could tell the businesses that really needed it already use these tools and the other businesses just don't have enough business dependent on optimization to be worthwhile.

cjalmeida 3y ago

IMO, the main benefit of those exact algos are the they provide optimality bounds, allowing you to make informed decision to continue or stop and provide a non-optimal solution. This is valuable in many practical situations. >Time scheduling, logistics, operations... Add marketing (pricing, assortment, ...) and finance (portfolio, captial budgeting, ...) to those. And those are huge domains, moving trillions of USD globally each year. Many companies (think most of the Global 2000) benefit from a couple % improvement on their operations that lead to many $$ in savings.

eru 3y ago
Even just arranging UI elements around a page could be done with these solvers, instead of coding up heuristics to satisfy your constraints by hand.

kj4211cash 3y ago

I don't think this is true. It's just that there aren't many optimization practitioners and the OR community has done a terrible job promoting itself. The value in scheduling and logistics is well established. So optimizers tend to congregate. But optimization could be applied to lots of other domains successfully. ML is just minimizing a loss function. I see simulation and post-hoc analyses being applied in numerous situations where optimization would be more efficient and produce a guaranteed optimal result. RL on simple, static problems is really an MDP but I've seen organizations fail to realize this. Etc.

Karrot_Kream 3y ago
How often do these come up? I studied some OR stuff in grad school when I was evaluating solvers and brushed up against the OR tools community for some research ideas and from what I could tell the businesses that really needed it already use these tools and the other businesses just don't have enough business dependent on optimization to be worthwhile.