On behalf of all those who are particularly concerned about this issue (I am not - particularly), this is very well put. The divisions between "code", "method" and "parameterization" are naturally very murky, and at the end of the day one probably has to apply a good deal of judgement and reason. I think the fear some people have is that, if things are not made explicit to the letter (presently, despite noble efforts, they are not), it could end up being a *court* - guided by HPN's formidable legal team - who decide what is "reasonable", after the event.

Personally, I don't see any ambiguity here - the rules don't use terms like "code", "method" or "parameterization", so there's no need to worry about such terms. It uses the words "particular" and "total", which seem to me to be fairly unambiguous - if you make any change, it's not the same "particular" algorithm any more, or if you remove any piece, it's not the "total" algorithm any more. I don't see that there's anything more HPN could write in the rules, or that I can say, to alleviate this concern. For anyone who is particularly concerned still, I can only suggest you talk to your own lawyer about how own particular situation.

Well, I don't want to be a stick in the mud, but as I understand the definition of "Prediction Algorithm",

"Prediction Algorithm" is the algorithm used to produce the data in an Entry taken as a whole (i.e., its particular total configuration) but does not include individual components of the Prediction Algorithm or tools used for analysis or development of the Prediction Algorithm.

...sounds like a program that has your entry stored in it and prints it out verbatim would qualify as a "Prediction Algorithm". I.e.,

#include <stdio>

char** data = { "12345,0.2", "12346,0" // and so on for 130k members

void main() {

for (int i = 0; i < members; ++i) std::cout << data[i] << std::endl;

}

I know this isn't what HHP wants and I'd expect to be disqualified if I submitted that as my "algorithm", but it follows the letter of the rules.

Tom SF Haines makes a good point with "The description being presented makes it sound like slightly tweaking even a single parameter circumvents exclusivity, which would make the use of the term irrelevant." If the definition of "Prediction Algorithm" is as narrow as Jeremy Howard says it is ("if you make any change, it's not the same "particular" algorithm any more, or if you remove any piece, it's not the "total" algorithm any more"), then what is the point of an exclusive license? Jeremy says "I don't see that there's anything more HPN could write in the rules, or that I can say, to alleviate this concern." Well, HPN could just remove the exclusivity provision, and that would probably alleviate most of the concerns expressed in this and related threads. -- Dave Slate

 

It's clear from Jeremy's post, and the language of the rules, that whatever exclusivity HPN is claiming is thin, and so I'm not sure what the "concern" is from the competitor side. Unless it's concern for the livelihood of HPN.

 

For me, feature selection seems like a bigger concern.

Like Cyfarwyddyd, I would prefer to concentrate on matters like feature selection rather than legal issues, but I also need to be careful about the meaning of the legal agreements I sign. Note that the FAQ says: "Why is HPN seeking to control my Prediction Algorithm? HPN wants to help ensure that Prediction Algorithms are used to promote better patient health care and not for improper purposes." This brings up the obvious question: what "improper purposes" could a "Prediction Algorithm" be used for that a minor revision of it (and therefore not subject to the exclusive license) could not? If the implied exclusivity is so thin, then why bother to claim it at all? -- Dave Slate

@Jeremy. I think part of the problem people now have is that the interpretation you indicate there would give HPN absolutely nothing... so much so that it can't possibly be the case in reality (as it can be interpreted that the smallest, almost superficial, of changes, lends a loophole to the legal obligations). I would beg to differ and suggest that the terms *could* be defined much better actually. However, no doubt not everybody would not be content with the result of that either, and you might end up going backwards and forwards for a long time. Realistically, I think people are going to have to go on faith. In everybody's favour, there has been considerable discussion on this board, by representatives of the competition, which I assume would be eligible as evidence in any very very unlikely legal case.

I would also add this is not just about court cases (Though such a case involving code you depend on to do your job, as would be the case with me, could seriously damage, or even end, your career.) but uncertainty. Many of us are going to be in the situation where code we currently have, and would like to use for the competition, and code we develop for the competition is an asset, which can be sold, traded or used for our own benefit - any uncertainty over the ownership of the code can be a serious problem, especially if you want to sell it.

Ok, one more point: Given that the best algorithms are going to be published (In the rules.) and that the submitter must have the legal right to submit it (i.e. it can't be covered by patents, not that software patents count in most countries anyway.) the claimed objective of exclusivity in the faq is absurd - under those conditions anyone can reimplement the algorithm and use it for whatever they want.

People complaining about HHP hurting their career... I think it boosts your career greatly if you end up high in ranks of this competition. It will be worth more than the money prize in this competition. It can be the fact that my English isn't good enough to understand the basics of the milestone prizes, but it seems to me that you don't want to win those prizes if you have to give insights in your methodology.

Not sure why I became the most vocal person in this thread - I certainly don't want to boycott this competition as it has some great potential, and indeed I would very much like to throw some non-parametric Bayesian graphical models at it. But solving problems such as this one is my job, and that means that I can't take any risk that entering this competition will negatively affect my ability to do that job, nor can I risk legal action, as I simply cant afford it. Whilst your right that the potential exposure of doing well could be highly beneficial the thing is that in academia coming 10th place in a competition such as this is of zero benefit if you can't then publish your algorithms and build on that research - right now the rules put both activities in a legal grey zone. Quite simply, the rules at present leave me hanging with regard to the primary reason I want to enter this competition, and the organisers do not even seem to realise that they have a problem.

As far as I understand it is restricted to publish algorithms involving the HHP data or created with the HHP data.

Bayesian models are much much more generally usable than only for this HHP competition. But if you would have a bayesian belief network producing usable results, using nodes/neurons assembled using the HHP data, the restriction applies. I totaly understand that.

I'm not an expert on laws in the USA, but in the Netherlands I will always be able to do my job. Regardless of some signed document.

At my last job I had to sign a document telling me that I could never take a job at an other company involving my speciality. I signed it because the governmental laws would override this document anyway.

In the most extreme case: Your algoritm uses adding and subtracting, so you can't add or subtract the rest of your carreer? That's not what this is about. The HHP is concerned that the data they provide or algorithms on this data will be used by other companies since the data is freely available.

Will HPN and Kaggle own worldwide sales rights to every entry?

I asked this question at the start of the HPN / Kaggle competition.  According to the rules, it appears that HPN and Kaggle will own full rights to every entry in the competition -- including worldwide sales, with no compensation to the developer.

On 15 April, I received an email from Anthony Goldbloom (Kaggle), saying that he was going to check with HPN.  I did not receive any follow-up, however.

Fundamentally, it appears that the "License" section of the agreement would give HPN and/or Kaggle the unlimited right to sell any "algorithm" or software used by every competitor who joins the Heritage Health Prize competition, whether they win a prize or not.

I can see granting a license to HPN for their internal use, as the sponsor who funded the competition.  Asking every competitor to grant a free license that would allow HPN (and Kaggle) to sell our work worldwide -- with no royalties -- is something else.

Was this the intent of the agreement?  If not, it needs to be clarified.  Anyone who has done serious work in predictive analytics knows the value of their IP -- and will refuse to participate.

Please let me know if this issue has been clarified by HPN and/or Kaggle.

Mark 

Yes. they will own every single entry, regardless of whether you win anything.  They will own worldwide, exculsive rights, and furthermore, you won't even be able to use your work as a "jumping off" point, in order to further research in some other area.  The original rules, when the competition started, didn't have this issue; you could continue to develop and market your work as you saw fit.  The rules were changed unfortunately.  Specifically, Rule 21 deals with this.

I think FineLineSysDes is probably right. The lack of further responses by Kaggle in this regard mean they are clearly ignoring us, given they must be watching this conversation, therefore I can only conclude several possible reasons:

1) They don't think there is a problem. Given this is the most viewed thread and that the points being made are hardly lacking merit or seriousness I find it unlikely that they are this stupid.

2) They are trying to get clarification from HPN, but have forgotten to mention that to us. Again, not that stupid.

3) They have gone back to HPN, who have confirmed the rules as they stand, and they have then shut up because they hope people will not leave in the uncertainty, and realise that explicitly saying this could provide a quote if the tech media decide to make a story out of this.

Only the third option makes sense, which is a shame. They are effectively excluding all academics, the people best positioned to generate completely new approaches. To put down $3m+ for research and then deny the participation of people who do research professionally seems extremely short sighted, to put it mildly.

I would like to think that this is a communication issue, and that it can be fixed - we are effectively communicating with HPN via Kaggle, and if Kaggle were to clearly spell the issues and consequences out to HPN one would hope they would adjust the rules such that the professionals can participate. I am going to continue watching this thread, but until I see evidence that the rules are going to be changed I, and many others, can not participate.