4 ms·

Greedy algorithms suck compared to what? If the greedy algorithm still converges fastest and is closest to optimal compared to any other practical scheme, the p

Greedy algorithms suck compared to what? If the greedy algorithm still converges fastest and is closest to optimal compared to any other practical scheme, the paper's result still does not say anything of practical use.

klyrs 6y ago
They suck at escaping local minima, for one. Your "if" is doing a ton of work there, and this paper is quite abstract market theory. The "practical use" is to better contextualize the efficient market hypothesis.
- unishark 6y ago
  They don't escape local minimum at all. And that means they suck for society at large compared to what? Random guessing? A corrupt government price system? An omniscient one that can solve NP hard problems? The "if" motivates answering the question. Anyway markets have a lot more going on than a simple greedy algorithm. For injecting noise into the optimization to get out of local minima. Markets clearly have a lot of this.
  - jabl 6y ago
    > And that means they suck for society at large compared to what? Random guessing? A corrupt government price system? An omniscient one that can solve NP hard problems? There's an intriguing recent book (that, admittedly, I haven't read) called "The People's Republic of Walmart" https://www.penguinrandomhouse.com/books/564287/the-peoples-republic-of-walmart-by-leigh-philips-and-michal-rozworski/ https://www.penguinrandomhouse.com/books/564287/the-peoples-... If large corporations are the shining beacons of capitalism, how come they are run internally as top-down planned economies? And to the extent they prove than planning can work at scale, could planning work society-wide? Not saying I agree or disagree (again, I haven't actually read it), but dusting off the ye olde calculation debate and updating it for today seems like an interesting idea.
    - unishark 6y ago
      I'd say startups are a much better example. But I didn't come here to argue politics, just asked for someone to describe their superior optimization process (math being my actual interest, so still waiting to hear it). I wouldn't describe "do what walmart did" as a particularly useful optimization strategy.
      - klyrs 6y ago
        > But I didn't come here to argue politics, just asked for someone to describe their superior optimization process Look. I don't believe that P=NP. I can't describe a superior optimization process, and if I did, I would claim the Clay Math prize. I'm also not an economist. My takeaway is that any simple strategy is almost certainly bad in the long term. You can optimize your own personal take, but in doing so, "the invisible hand" isn't going to somehow flip that into a benefit for all.
    - imtringued 6y ago
      The worst part of centrally planned economies is that they are prescriptive and essentially authoritarian in nature. If they make mistakes then nothing prevents them from repeating the same mistake. There is no threat of the central planner being replaced. If all Walmart does is make mistakes then they go out of business. Central planning won't work but federated planning with 20 super corporations that are in competition with each other might actually be good enough. The only problem is that we both need and want the underdogs(startups) and this would undermine them.
      - rbanffy 6y ago
        > There is no threat of the central planner being replaced You are conflating an economic model with a political one. In fact, the central planner doesn't even need to be a person, but a set of assumptions and strategies that can be measured against the benefit they provide and be continually adjusted. You touch an important point, however. The people most affected by Walmart's central planning, their employees, have no say in what strategies and assumptions the central planner uses and that removes any incentive to provide benefit to them.
mcbits 6y ago
Greedy algorithms can be arbitrarily bad on large NP-hard problems. But if we're using terms metaphorically, we could also say each greedy company is just a Monte Carlo sample of the whole economy. Some fail or stagnate in local optima; others succeed for longer. But the economy is too complex to expect anyone to ever find a global optimum, so continuous exploration of the landscape is important for overall growth. Thus we can aim for and admire success but should be vigilant against anticompetitive behaviors that stifle the sampling.

klyrs 6y ago

They suck at escaping local minima, for one. Your "if" is doing a ton of work there, and this paper is quite abstract market theory. The "practical use" is to better contextualize the efficient market hypothesis.

unishark 6y ago
They don't escape local minimum at all. And that means they suck for society at large compared to what? Random guessing? A corrupt government price system? An omniscient one that can solve NP hard problems? The "if" motivates answering the question. Anyway markets have a lot more going on than a simple greedy algorithm. For injecting noise into the optimization to get out of local minima. Markets clearly have a lot of this.
- jabl 6y ago
  > And that means they suck for society at large compared to what? Random guessing? A corrupt government price system? An omniscient one that can solve NP hard problems? There's an intriguing recent book (that, admittedly, I haven't read) called "The People's Republic of Walmart" https://www.penguinrandomhouse.com/books/564287/the-peoples-republic-of-walmart-by-leigh-philips-and-michal-rozworski/ https://www.penguinrandomhouse.com/books/564287/the-peoples-... If large corporations are the shining beacons of capitalism, how come they are run internally as top-down planned economies? And to the extent they prove than planning can work at scale, could planning work society-wide? Not saying I agree or disagree (again, I haven't actually read it), but dusting off the ye olde calculation debate and updating it for today seems like an interesting idea.
  - unishark 6y ago
    I'd say startups are a much better example. But I didn't come here to argue politics, just asked for someone to describe their superior optimization process (math being my actual interest, so still waiting to hear it). I wouldn't describe "do what walmart did" as a particularly useful optimization strategy.
    - klyrs 6y ago
      > But I didn't come here to argue politics, just asked for someone to describe their superior optimization process Look. I don't believe that P=NP. I can't describe a superior optimization process, and if I did, I would claim the Clay Math prize. I'm also not an economist. My takeaway is that any simple strategy is almost certainly bad in the long term. You can optimize your own personal take, but in doing so, "the invisible hand" isn't going to somehow flip that into a benefit for all.
  - imtringued 6y ago
    The worst part of centrally planned economies is that they are prescriptive and essentially authoritarian in nature. If they make mistakes then nothing prevents them from repeating the same mistake. There is no threat of the central planner being replaced. If all Walmart does is make mistakes then they go out of business. Central planning won't work but federated planning with 20 super corporations that are in competition with each other might actually be good enough. The only problem is that we both need and want the underdogs(startups) and this would undermine them.
    - rbanffy 6y ago
      > There is no threat of the central planner being replaced You are conflating an economic model with a political one. In fact, the central planner doesn't even need to be a person, but a set of assumptions and strategies that can be measured against the benefit they provide and be continually adjusted. You touch an important point, however. The people most affected by Walmart's central planning, their employees, have no say in what strategies and assumptions the central planner uses and that removes any incentive to provide benefit to them.

mcbits 6y ago

Greedy algorithms can be arbitrarily bad on large NP-hard problems. But if we're using terms metaphorically, we could also say each greedy company is just a Monte Carlo sample of the whole economy. Some fail or stagnate in local optima; others succeed for longer. But the economy is too complex to expect anyone to ever find a global optimum, so continuous exploration of the landscape is important for overall growth. Thus we can aim for and admire success but should be vigilant against anticompetitive behaviors that stifle the sampling.