I think it would show a kind of gilded age maturity if all the cloud providers cooperated on their public datasets, because they are for the public good.
Hacker News is a member of this dataset, but it's only updated daily. If you want to get meta about this discussion, I guess you'll need to wait until tomorrow.
This seems critical - as Google is a corporation, I wonder where this sense comes from? Obviously we have certain expectations for a couple who's tagline reads "don't be evil," but does that extend to criticizing them for not providing terabytes of free transfers of terabytes of free data, which they pay to host and serve?
> criticizing them for not providing terabytes of free transfers of terabytes of free data
My criticism is not towards them not providing free transfers of data. My criticism is that they say "these datasets are freely hosted and accessible" and public, while only being available while logged in to Google, which I don't think classifies as public.
I would not have any problem if it's just called "Google-hosted Datasets" or "Google-only Public Datasets", I just think the current naming is misleading.
the term public means its available to the public. Signing up google, library card, doesn't negate that fact that any member of the public can access the data with a reasonable and insignificant barrier.
Sure, the difference between getting a library card and signing up to a Google account is that usually the first has a user agreement which is about 10 lines long with human language and the second one has a user agreement which is 10+ pages with lawyer speak and probably includes that they are allowed to sell your data. How is it reasonable for a normal person to understand those kind of user agreements?
And if that's your definition of public, isn't everything public, if you have the right amount of money, knows the right people and can get access to the right place?
You only pay a price because you are using someone else system. Which cost money. Public library are public, you also pay taxes on them. Because using it as a service still has a cost.
Google give you 1TB on data to query each month. That is a lot of data, if you need more for free collect the data over time.
But you still need a library card and if the facility gets overused they ask for a small tax to cover the operating costs.
I fail to see how this isn't considered public. Public literally just means accessible to the general public. That all. A public event can still charge a fee for entry. As long as the fee doesn't create a barrier for most the public, you know accessible to the general public.
Also if a library sold their data would it no longer be considered public? I believe it would still be considered public. User data isn't sold typically by name or email, byt activity. If a library compiled a list of books checked out and their frequency, the amount of people entering everyday etc that be the equivalent of most user data being sold. Very rare for a company to sell your actual personal data, when they do they disassociate your personal information with it.
So if the above doesn't disqualify a library from being public, then neither would this dataset that is public. If you really disagree with that then you are just trying to be pedantic at that point.
If anybody is interested in AI/bioinformatics projects on AWS, I'm currently involved in a project to harmonize _all_ of the publicly available RNA data (many petabytes) into easily sliced, AI/ML-ready datasets for anybody to use:
An RNA sequencing run generates on the order of 10GB of data, a typical study requires many runs (treatments, controls, replication of results, etc), and posting the raw data is required by most biology journals. I'm not surprised that there is over 1PB of data available to curate.
I'm talking about the raw reads, which is important if you want to try a different alignment or base-calling method. You can debate how important it is to be able to do that, but I'm not trying to argue that the data should be kept, I was just explaining why the total size of publicly available RNA-seq data (the sum total of which the parent is attempting to organize) runs in the petabytes.
So, do you or the original poster actually have a materialized petabyte of RNA data? Otherwise, you're just describing a million files spread over a million file servers, not being used for science or processed in any way.
As someone who has spent a considerable amount of time on data that has ended up on this page, I think the fact that the list hasn't grown says more about the priorities of other companies than of AWS. Amazon doesn't (yet) have time to build and maintain these datasets themselves: they work with others to build and maintain it and then fund the storage and transmission fees.
I helped build the Terrain Tiles dataset as part of Mapzen, which recently shut down. The OpenStreetMap data exists on the AWS Public Datasets page because it's useful to Humanitarian OpenStreetMap Team. If you're able to convince your company to generate and work with a public dataset, consider reaching out to the AWS and Google public datasets teams to get it hosted and publicized.
In my experience, the list that AWS keeps on their public page isn't completely up to date. There are two major datasets in the neuroimaging community that are hosted in S3 (see my comment elsewhere on this page), but AWS hasn't widely publicized this fact for some reason.
How much does it cost to export/process this data from Amazon? Unlike GCP/BigQuery which has free cloud processing built in, downloading/analyzing these GB/TB datasets for personal analysis can’t be cheap.
The descriptions note “Educators, researchers and students can apply for free promotional credits to take advantage of Public Datasets on AWS.” which is not a good sign.
It's the normal data cost. EG the Landsat data is stored in S3, so it's free to EC2 (in the same region), and somewhere around $0.09/gb to the public internet.
However, the idea is not that you download it all (there's probably cheaper ways to acquire Landsat data), the idea is if you want to do whatever analysis on AWS, they've already got it neatly ingested for you.
Importance of data for machine learning algorithms can’t be stressed enough. I remember talking to a friend in Google Translate team. they had a good also but they were struggling to get quality translation data to train their service. the problem was more severe when it came to language which was not very popular. translation set was next to nothing for Say something like Turkish, Hindi, Latvian etc.
They finally solved this by using meeting notes from UN Assembly. Which were transcribed by best of the translators? that access to meeting transcription was (unfair ??) advantage Google had over other tools. Was it wrong ? I don’t think so. Should have been those meeting notes be public: Yes
When Google Translate was new in my language Finnish, their translation quality was atrocious.
I tried this sentence: "Hei äiti, puhun suomea". The expected translation would be "Hi mom, I speak Finnish".
Instead Google's result was: "July's mother, I speak English".
Obviously the engine had been trained on unvetted data sets where the word "English" occurred in translations in a position where the original had the word "Finnish", and no context was provided to avoid this kind of mistake.
The word "July" came about because "hei" is also used as an abbreviation for "heinäkuu" (July). It was sobering that a supposed world-class AI couldn't distinguish between these two usages. Machine learning needs a lot of old-fashioned handtuned human-made heuristics.
The German ML translation company of ex-googlers called DeepL (formerly Linguee) providing the currently best machine translation service apparently got a lot of their data from (I assume public) EU documents which had to be translated to all member languages.
This doesn't appear to be publicized on the linked to page, but two of the major open access fMRI dataset repositories (FCP-INDI [1] and OpenfMRI [2] are hosted in S3.
You can get the URL data (and parse out the domains yourself) at http://index.commoncrawl.org/
Be aware that commoncrawl, while huge, only crawls a fraction of the web, and so you won't get all of the "existing domains" (i.e. all domains that exist) if that is what you are after.
I think it would show a kind of gilded age maturity if all the cloud providers cooperated on their public datasets, because they are for the public good.