5 ms·

I'm officially calling bullshit on the algorithm requiring "one quintillion variables".

by scovetta 14y ago

sjwright 14y ago
Perhaps they're keeping track of each atom individually.
tantalor 14y ago
Combinatorial explosion?
unwind 14y ago
I was skeptical too, but all it takes is 60 bits, i.e. 60 "yes/no"-questions in their "black book". Doesn't sound too far-fetched. But, of course, I do hope they're not brute-forcing it. See http://www.wolframalpha.com/input/?i=log2%28one+quintillion%29 http://www.wolframalpha.com/input/?i=log2%28one+quintillion%....
- kd5bjo 14y ago
  That's only 60 variables, which can produce a quintillion variations.
mturmon 14y ago
I think bullshit is it. The best way I could think of to generate lots of variables is a random forest classifier, which is an ensemble of (say) thousands of trees. Each tree constructs a lot of variables at each node, so maybe you'd have (stretching here) hundreds of thousands to perhaps millions of variables. But 1e6 << 1e18. Maybe they're counting each pixel of some remote sensing image of orange groves? Bullshit again, that would have to be a 1e9 x 1e9 pixel image. I'd believe it from NGA (http://en.wikipedia.org/wiki/National_Geospatial-Intelligence_Agency http://en.wikipedia.org/wiki/National_Geospatial-Intelligenc...), but not from Coke. If there's a shred of truth to it, they're using some dodgy mechanism to increase the variable count.
- greatzebu 14y ago
  It seems more likely that the reporter made a mistake, and there are actually a quintillion possible internal states generated from a relatively small number of variables, not a quintillion variables.
mistermcgruff 14y ago
BlackBook is a mixed-integer programming model implemented in OPL/CPLEX (used to be an ILOG product until IBM bought them). So all this stuff about classifiers is off base. This isn't a predictive model -- it's a math program. I don't know how they got that quintillion number. Could be that many "variations" instead. But the dimensions are things like time, varietals, locations, what you're gonna store/blend, that kind of thing. You end up with decision variables that are indexed in multiple dimensions and that creates the variable explosion they're alluding to. CPLEX is awesome at handling huge amounts of decision variables so long as the problem stays mixed-integer, and that's what this is.

I'm officially calling bullshit on the algorithm requiring "one quintillion variables".

sjwright 14y ago

Perhaps they're keeping track of each atom individually.

tantalor 14y ago

Combinatorial explosion?

unwind 14y ago

I was skeptical too, but all it takes is 60 bits, i.e. 60 "yes/no"-questions in their "black book". Doesn't sound too far-fetched. But, of course, I do hope they're not brute-forcing it. See http://www.wolframalpha.com/input/?i=log2%28one+quintillion%29 http://www.wolframalpha.com/input/?i=log2%28one+quintillion%....

kd5bjo 14y ago
That's only 60 variables, which can produce a quintillion variations.

mturmon 14y ago

I think bullshit is it. The best way I could think of to generate lots of variables is a random forest classifier, which is an ensemble of (say) thousands of trees. Each tree constructs a lot of variables at each node, so maybe you'd have (stretching here) hundreds of thousands to perhaps millions of variables. But 1e6 << 1e18. Maybe they're counting each pixel of some remote sensing image of orange groves? Bullshit again, that would have to be a 1e9 x 1e9 pixel image. I'd believe it from NGA (http://en.wikipedia.org/wiki/National_Geospatial-Intelligence_Agency http://en.wikipedia.org/wiki/National_Geospatial-Intelligenc...), but not from Coke. If there's a shred of truth to it, they're using some dodgy mechanism to increase the variable count.

greatzebu 14y ago
It seems more likely that the reporter made a mistake, and there are actually a quintillion possible internal states generated from a relatively small number of variables, not a quintillion variables.

mistermcgruff 14y ago

BlackBook is a mixed-integer programming model implemented in OPL/CPLEX (used to be an ILOG product until IBM bought them). So all this stuff about classifiers is off base. This isn't a predictive model -- it's a math program. I don't know how they got that quintillion number. Could be that many "variations" instead. But the dimensions are things like time, varietals, locations, what you're gonna store/blend, that kind of thing. You end up with decision variables that are indexed in multiple dimensions and that creates the variable explosion they're alluding to. CPLEX is awesome at handling huge amounts of decision variables so long as the problem stays mixed-integer, and that's what this is.