5 ms·

While the Kaggle competitions are cool for students, 50K would be a measly amount paid to an R&D company or contractor to do such a job. I don't like the bounty

While the Kaggle competitions are cool for students, 50K would be a measly amount paid to an R&D company or contractor to do such a job. I don't like the bounty approach to data science the same way I don't like it for software engineering.

foota 9y ago
Sure, it's not much for well compensated contractors, but 50k can be a lot of money for some people.
- stingrae 9y ago
  The point is that with 1000 people working on it, 999 are working for free.
  - onion2k 9y ago
    There are tens of thousands of people working on code to do object recognition for free already. This is just a incentive to get them to tackle a particular problem in a vast sea (heh!) of problems they could work on. It's quite likely that very few people understand that this even is a problem, and fewer still who'd be interested in it enough to work on it. This isn't getting people to work for free. It's simply getting people interested in a problem.
    - craftyguy 9y ago
      From their perspective, you are doing work for them, for free. What you just described is a benefit you receive (or more accurately, perceive) but ultimately cost them essentially $0.00 to provide.
  - paulgb 9y ago
    I spent a couple hours building a CNN for an entry to this competition last night. I don't consider it working for free, I consider it a way to practice on an interesting labeled dataset I'd otherwise not have access to.
  - whataretensors 9y ago
    This is a big reason why I don't do kaggle contests, especially when the solution is potentially worth a lot of money.
Simon_says 9y ago
I guess they're not trying to hire an R&D company or a contractor, then.
autokad 9y ago
as far as i know, most kagglers are not students, but honestly i havent seen any data on that. I'll speak for myself, I do it because I love trying to solve interesting problems and I love data science. thus far every competition I joined I learned something new and expanded my ds code base. I also really love the fact people post kernels and discussions I can learn from as well.
- joshvm 9y ago
  It's very difficult to win a Kaggle comp these days, not because of technical difficulties, but because you're up against teams of people with the resources to try a lot more hyperparameters than you. To a certain extent you can win with money - if you can afford to spend money on a bunch of AWS instances, that gives you more iteration time than one person with a 1080 in their bedroom. (So no, they're mostly not students.) On the other hand the problem sets are 'real' and it's excellent machine learning experience - importanly because you have the competition element, you get an idea about how good your solution actually is.
  - autokad 9y ago
    in the zillow one, I was 41/3800, but I got over-obsessed with lb results and the first round of home sales my rank droped to 118. however, this was my first real kaggle competition, and I only spent a few weeks on it, (I'd say about 3 total). it was also one of the more competitive competitions, with a million dollars on the line. Its exponentially difficult to move up the ranks, but I do think its realistic to win one (though hard - as it should be). a lot of time, effort, and skills required but not impossibly difficult. I think kaggle would do everyone a favor if they gave out awards to more people, instead of the winner take all mentality. at the very least, everyone in the top 10 should win money, not just the top 3, with 1 getting the lions share.
HarryHirsch 9y ago
With that kind of bounty you aim to attract someone who has already solved the problem and has the solution sitting in his drawer. Eli Lilly pioneered the crowdsourcing approach in 2001 with their spinout Innocentive, and they raised eyebrows even then because they were perceived as cheap.
Blazespinnaker 9y ago
Think it’s bad now, just wait for the singularity.

foota 9y ago

Sure, it's not much for well compensated contractors, but 50k can be a lot of money for some people.

stingrae 9y ago
The point is that with 1000 people working on it, 999 are working for free.
- onion2k 9y ago
  There are tens of thousands of people working on code to do object recognition for free already. This is just a incentive to get them to tackle a particular problem in a vast sea (heh!) of problems they could work on. It's quite likely that very few people understand that this even is a problem, and fewer still who'd be interested in it enough to work on it. This isn't getting people to work for free. It's simply getting people interested in a problem.
  - craftyguy 9y ago
    From their perspective, you are doing work for them, for free. What you just described is a benefit you receive (or more accurately, perceive) but ultimately cost them essentially $0.00 to provide.
- paulgb 9y ago
  I spent a couple hours building a CNN for an entry to this competition last night. I don't consider it working for free, I consider it a way to practice on an interesting labeled dataset I'd otherwise not have access to.
- whataretensors 9y ago
  This is a big reason why I don't do kaggle contests, especially when the solution is potentially worth a lot of money.

Simon_says 9y ago

I guess they're not trying to hire an R&D company or a contractor, then.

autokad 9y ago

as far as i know, most kagglers are not students, but honestly i havent seen any data on that. I'll speak for myself, I do it because I love trying to solve interesting problems and I love data science. thus far every competition I joined I learned something new and expanded my ds code base. I also really love the fact people post kernels and discussions I can learn from as well.

joshvm 9y ago
It's very difficult to win a Kaggle comp these days, not because of technical difficulties, but because you're up against teams of people with the resources to try a lot more hyperparameters than you. To a certain extent you can win with money - if you can afford to spend money on a bunch of AWS instances, that gives you more iteration time than one person with a 1080 in their bedroom. (So no, they're mostly not students.) On the other hand the problem sets are 'real' and it's excellent machine learning experience - importanly because you have the competition element, you get an idea about how good your solution actually is.
- autokad 9y ago
  in the zillow one, I was 41/3800, but I got over-obsessed with lb results and the first round of home sales my rank droped to 118. however, this was my first real kaggle competition, and I only spent a few weeks on it, (I'd say about 3 total). it was also one of the more competitive competitions, with a million dollars on the line. Its exponentially difficult to move up the ranks, but I do think its realistic to win one (though hard - as it should be). a lot of time, effort, and skills required but not impossibly difficult. I think kaggle would do everyone a favor if they gave out awards to more people, instead of the winner take all mentality. at the very least, everyone in the top 10 should win money, not just the top 3, with 1 getting the lions share.

HarryHirsch 9y ago

With that kind of bounty you aim to attract someone who has already solved the problem and has the solution sitting in his drawer. Eli Lilly pioneered the crowdsourcing approach in 2001 with their spinout Innocentive, and they raised eyebrows even then because they were perceived as cheap.

Blazespinnaker 9y ago

Think it’s bad now, just wait for the singularity.