Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Open Source Datasets

deepmind.com

Post reply on HN
Print view View original

21–29 of 29 posts

Re: Open Source Datasets

#21

Post by zitterbewegung »

nerdponx wrote: post #18
zitterbewegung wrote: post #4

I have been thinking about what this solves in respect to other datasets. Nearly all shape recognition datasets have a restriction that you can't use unless you are an academic. I feel like that Open Sourcing data sets will allow us to be more democratic with data and the things that are generated by them. Creative Commons seems like a good license for this though. Once you have the data is half the battle . The rest…

Nearly all shape recognition datasets have a restriction that you can't use unless you are an academic. I feel like that Open Sourcing data sets will allow us to be more democratic with data Just because something is open source doesn't mean it can't have an academia-only restriction. Data sets should cost money in for-profit uses, open or not.

I agree that people who use a dataset for profit should be required for a for profit use. I didn't mean to imply that academics should give their data away for free. Maybe there needs to be a way for academics to sell their data or any other tangible benefit.

Re: Open Source Datasets

#22

Post by ptero »

nerdponx wrote: post #17
ptero wrote: post #15

One question that is not clear to me is what should the dataset license to allow / restrict, in the perfect world. For me (just a personal opinion) it would allow free (as in liberty) use, but somehow encourage those who use it to share the benefits (data, software or algorithms) under the same license. Unfortunately, Open Source does not help here -- I do not see how OS can be used with data sets. The main OS levera…

OS helps tremendously in reproducibility. Without the underlying data, there is no way to audit an analysis. Moreover, an algorithm is only ever "done" in the same way that software is ever "done". New techniques might come along that could enhance the model, or the business requirements might change that necessitate re-tuning the algorithm. The benefits of OS data are the same as the benefits of OS software. The dis…

Maybe I was not clear -- I am not arguing for proprietary datasets. Validation on publicly available data is a key component in comparisons, assessments, etc.

However, there is a whole bestiary of open source licenses that span the spectrum of "use any way you want" to much more restrictive. But they were mostly thought through for software and data is different; what may prevent proprietary abuse in software may not have any teeth for data.

Re: Open Source Datasets

#23

Post by tuukkah »

zitterbewegung wrote: post #4

I have been thinking about what this solves in respect to other datasets. Nearly all shape recognition datasets have a restriction that you can't use unless you are an academic. I feel like that Open Sourcing data sets will allow us to be more democratic with data and the things that are generated by them. Creative Commons seems like a good license for this though. Once you have the data is half the battle . The rest…

Open data is a concept separate from free and open source software. However, they have similar concerns such as copyleft: GPL for source code and ODbL for data both require changes to be published.

Outside of copyleft, Creative Commons 4.0 licenses (CC-BY 4.0, CC-BY-SA 4.0, CC0) from 2013 are good licenses for open data: https://theodi.org/blog/cc-40-and-open-data

Re: Open Source Datasets

#24

Post by tuukkah »

nerdponx wrote: post #18
zitterbewegung wrote: post #4

I have been thinking about what this solves in respect to other datasets. Nearly all shape recognition datasets have a restriction that you can't use unless you are an academic. I feel like that Open Sourcing data sets will allow us to be more democratic with data and the things that are generated by them. Creative Commons seems like a good license for this though. Once you have the data is half the battle . The rest…

Nearly all shape recognition datasets have a restriction that you can't use unless you are an academic. I feel like that Open Sourcing data sets will allow us to be more democratic with data Just because something is open source doesn't mean it can't have an academia-only restriction. Data sets should cost money in for-profit uses, open or not.

“Open data and content can be freely used, modified, and shared by anyone for any purpose” http://opendefinition.org/

"The license must not restrict anyone from making use of the program in a specific field of endeavor. For example, it may not restrict the program from being used in a business, or from being used for genetic research." https://opensource.org/osd

Re: Open Source Datasets

#26

Post by jackschultz »

This brings up a huge point about how important data sets are to analysis and machine learning. There are so many libraries out there that make learning algorithms quick to run, and the absolute most important part of a project of that type is correct and formatted data.

Re: Open Source Datasets

#27

Post by guidopallemans »

Abtin88 wrote: post #12
Kpourdeilami wrote: post #2

Somewhat unrelated: Deepmind's website is so cluttered and distracting to the extent that it is almost unusable

reminds me of the motherfucking website! http://motherfuckingwebsite.com/

http://bettermotherfuckingwebsite.com/

I actually use their css when formatting my .md files

Re: Open Source Datasets

#28

Post by nerdponx »

tuukkah wrote: post #24
nerdponx wrote: post #18

Earlier quoted context omitted.

Nearly all shape recognition datasets have a restriction that you can't use unless you are an academic. I feel like that Open Sourcing data sets will allow us to be more democratic with data Just because something is open source doesn't mean it can't have an academia-only restriction. Data sets should cost money in for-profit uses, open or not.

“Open data and content can be freely used, modified, and shared by anyone for any purpose” http://opendefinition.org/ "The license must not restrict anyone from making use of the program in a specific field of endeavor. For example, it may not restrict the program from being used in a business, or from being used for genetic research." https://opensource.org/osd

Interesting. So is a nominally open-source program like RStudio not actually open-source because enterprise uses need a license?

Re: Open Source Datasets

#29

Post by SloopJon »

nerdponx wrote: post #28
tuukkah wrote: post #24

Earlier quoted context omitted.

“Open data and content can be freely used, modified, and shared by anyone for any purpose” http://opendefinition.org/ "The license must not restrict anyone from making use of the program in a specific field of endeavor. For example, it may not restrict the program from being used in a business, or from being used for genetic research." https://opensource.org/osd

Interesting. So is a nominally open-source program like RStudio not actually open-source because enterprise uses need a license?

It looks like RStudio is AGPL v3, which is a free and open source license. Enterprise users only need a different license if they don't want to abide by the AGPL's strong copyleft requirements.
Post reply on HN