Hacker News new | comments | show | ask | jobs | submit login
Announcing the Netflix Cloud Prize (github.com)
78 points by jedberg 1566 days ago | hide | past | web | 43 comments | favorite

Some of the evaluation criteria [1]:

* Submission provides evidence that code is in use by other projects, or is running in production at Netflix or elsewhere

* Submission has a large number of watchers, stars and forks on github.com

So the entrant has to prove that it is used in production at Netflix? And why would they care about the number of watchers/stars/forks? If I made a great contribution to something very specific (let's say aiming for the "Best contribution to performance improvements" or "Best contribution to code quality" awards), why would anyone watch/star/fork my repo? This seems more like a popularity contest to be honest...

[1] https://github.com/Netflix/Cloud-Prize/wiki

Coming up with measurable criteria is hard. It may not be perfect, but how else would you gauge community adoption?

Or are you complaining about netflix wanting projects "that contain a substantive contribution, working code and which have high community adoption"?

If that's what they want, then that's what they want and should award prizes accordingly.

Good point.

I guess I'm partly missing the point of the competition, and missed the fact that the entry could be anything(?) and does not have to do with something directly Netflix-related (either customer facing or operations).

I am not going to debate that. As a long time Netflix customer, the recommendation engine does not impress me. This is what I was recently recommended (thriller? Okay), see anything wrong with this picture:


Hah. If you look for the animated Pixar movie "Brave" and look at the "More like Brave" section it suggests "Magic Mike" (Yeah the stripper dude movie). That algorithm needs work.

"Or elsewhere"

The contest is exactly a popularity in most of those categories. Make something people find useful and share it. That's the entire idea behind OSS.

I imagine the performance categories are going to be judged a little bit differently, but for most categories the criteria look reasonable to me.

Looks like their definition of useful is based on how many people actually use it, by starring/forking the repo.

I'm sure this will be massively popular, but the cynic in me can't help but think this is a great way for a company to do very cheap R&D.

It's similar to their algorithm contest - where they basically had a fixed potential investment but reaped rewards thousands of times greater than what they could have afforded by tapping into people willing to work for free as a hobby.

On one hand I think it's a great idea, but on the other I'm concerned about how this can be exploited; industries built on passions tend to massively undervalue employees. See: Aviation, Sports, Publishing.

... cheap R&D and cheap publicity.

Much like the infamous recommendation data challenge, that was never put in production.

I am, too, a cynic - and I think their data model is flawed from the beginning (user account vs. household, among other things), causing schizophrenic recommendations.

The important aspects of the recommendation data challenge was put in production. Of course, it wasn't the precise weights and models, but the important ideas, namely ensembling and RBMs (which didn't exist before the challenge) were put in place. What you're describing is a misunderstanding of the statement from Netflix.

> user account vs. household

Profiles are being tested now and will soon be deployed to all customers.

Woo! Will there be a way to disentangle existing watch-instantly histories or to simply reset a profile with a clean one?

I don't think you can reset a profile, but the workaround is to just make a new profile, which will have a clean slate. Then you can go through and rate a couple things to help it learn.

What developer would participate in this after Netflix pulled the rug out on their API?

A developer that's noticed that they've open sourced some pretty useful and significant cloud-oriented projects?

Netflix didn't really pull the plug on the API -- everyone who was using it (which were only a few people) are still getting the same experience as they had before. There will just be no new growth in the use of the API.

Since a lot of us may be familiar with your old employer, it's probably worth noting that we're discussing your current employer.

Fair enough.

Disclaimer: I work for Netflix.

So, they want lots of engineers to do work for them and then only pay the one they like?

I was at their meetup yesterday. What you're implying is definitely _not_ the vibe.

I thought Netflix OSS is great. They're opensourcing something like 30 projects, putting on super well organized meetups with expo like rooms where you can pick Netflixers brain about the software and their internal systems in general, free food&beers, free Ts, etc. It was great.

They do it go give back to the community, to get patches and ports from the community, and hiring of course, there's nothing wrong with that!

I was there too. I went with a coworker who is a former Netflix employee. He was almost celebrating with each announcement. His biggest complaint about working at a startup is not having access to the tools they have built up. Netflix has built up an amazing platform that is tightly integrated with AWS. By making these tools open source our little startup has a quicker route to a robust and scalable platform. In my opinion that is awesome.

Or, they're encouraging developers to have fun with their open source projects and putting a prize at the end of it. It's not like there's someone twisting the ends of his moustache contemplating turning them closed source once they've got some good code on there.

The want lots of engineers to do open source development licensed under the Apache License. And they will pay 10 that they like.

After the Python at Netflix post, http://techblog.netflix.com/2013/03/python-at-netflix.html, I was expecting some Python repos on GitHub. It seems that all the open source projects they have use Java mainly. https://github.com/Netflix

Our first Python contribution is coming by next Tuesday, and that will be the bakery, aka Aminator.

More Python should follow shortly. Python is much newer here, so we aren't as far along the open source path.

Thats great to hear the ami bakery is finally getting open sourced. I wrote my own ami bakery and it has completely changed the way we do releases at my startup. I'm Looking forward to hacking on this for the contest.

You don't like traversing directories like deep nested ones here before you get to any code?


Is it just me or does the repo that we're supposed to be forking only contain the readme and the contest rules? Is there a baseline app to be extended?

The repo you fork is just a readme. Then you add in your submission, which you derive from all of the other open source software in the Netflix Github account.

Does anyone have any suggestions on how to make this clearer? I've seen this question a few times.

Edit: Thanks for the feedback everyone. I've incorporated it into the page.

Link to this more prominently: http://netflix.github.com/

Also maybe spitball some examples of what you expect people will do with this. Lots of cool repos there but it is a bit overwhelming.

So if you're adding tests to an existing repo do you have to copy the whole repo first instead of doing a pull request?

Put what you just said in the readme. With links. :-)

A link to the netflix github?

go up a level to get to their good stuff:


All the projects eligible for this contest are Java.... :\

Wait a couple of days and there will be a Python project.

You can also write a new monkey in any language you want.

So, for instance, I could develop an Edda client using C++11 and OpenGL for OSX?

Definitely! Sounds like a perfect entry for the portability prize.

They've done something like this with movie recommendations in 2006 -- got a super-cheap solution to a hard research problem.

That's not entirely fair. What Netflix got wasn't production ready, but some of the ideas are still in the current version of the recommender.

A quick read of this thread suggests that you could really get out in front of things with a well-placed FAQ, whether on a Netflix site, github, or your personal site.

Assuming your job description is something like relations.

Wow that's a cheap way to get free consultants. Assuming 40 new features get created... giving the best one a price. I kinda hate these competitions.

However, they are a fun way to recruit.

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | DMCA | Apply to YC | Contact