arXiv about archives

[jbenet]

👏 👏 👏 👏 @davidar on mirroring the CC part of arxiv on IPFS.

[rht]

Don't have enough network+storage bandwidth to do this, but wish to cat the source archives https://gateway.ipfs.io/ipfs/QmPFiQcfUPr9DYTJ7MPzY9ixxWkfK1oqQQEmQHS1zeD5PP and then process it to get the most used operators 1-gram / var names.

e.g. with 50 sources (from 1 field):

1. { 15425 590.54%
2. _ 9059 346.82%
3. ^ 7225 276.61%
4. - 4081 156.24%
5. + 3019 115.58%
6. = 2612 100.00%
7. \frac 1757 67.27%
8. \mu 1485 56.85%
9. \partial 1295 49.58%
10. \bar 1290 49.39%

If brevity of a notation comes with usage.

[davidar]

@rht what I'd really love to do is http://latexsearch.com

But more generally, running computations on large distributed datasets is an interesting problem. IPMR (map reduce) anyone? It would be cool if you could pay nodes (bitcoin?) to run some sandboxed code over their local blockstore and publish the results back to the ipfs network. @jbenet have you had any thoughts on this topic?

[davidar]

@rht Also, if you could give me the script to do this I'd be happy to run it on pollux

[rht]

(on latexsearch : it is too crude. it's like searching through latex "bytecodes" instead of the formulas it represent)

I will put the script here after I have modified it to walk through several files.
Currently it is https://ipfs.io/ipfs/QmQtLp5BxYVgo6miYz3EUAaoP32eqUcW6G8ywsiMxGB5vT

[davidar]

rht: I guess my interest is more academic, as in, has anybody written anything about this expression I've just derived?

Also, would be really cool to finalise this issue soon (and publicise it a little bit). @whyrusleeping what's the best way to proceed with adding lots of small files at the moment? Edit: use the tar adder? Are you able to do something like ipfs cat /ipfs/hash/archive.tar/foo/bar.txt?

Edit: and also do cool stuff with it to showcase the benefits of publishing with a creative commons license. CC-BY-SA is really the embodiment of the scientific social contract

[rht]

But it would be better if there are specifiers e.g. if the expression is vectorized, approximate/exact, etc.

https://ipfs.io/ipfs/QmcH6av29y7fXBqifmEMU54vVuMYjpBFBdcQsDkWZK8Hbi should work with python parselatex.py dirname.
Also wanted to know how long it takes to traverse the entire CC arxiv.

[davidar]

@rht definitely, and that's exactly why an open source version of LaTeX search would be better ☺

• run https://ipfs.io/ipfs/QmcH6av29y7fXBqifmEMU54vVuMYjpBFBdcQsDkWZK8Hbi

[davidar]

@rht https://ipfs.io/ipfs/QmeFkmYSPhn33hzEHxxqtvDNxHorW93iZFFMgZDQKrfeH4

real 6m24.104s
user 3m34.448s
sys 1m14.448s

[jbenet]

We should get @rht access on pollux -- @rht what's your ssh pubkey? 

—
Sent from Mailbox

On Tue, Sep 15, 2015 at 6:13 AM, David A Roberts [email protected]
wrote:

@rht https://ipfs.io/ipfs/QmeFkmYSPhn33hzEHxxqtvDNxHorW93iZFFMgZDQKrfeH4

real 6m24.104s
user 3m34.448s

sys 1m14.448s

Reply to this email directly or view it on GitHub:
#2 (comment)

[rht]

That's too fast!

This would have taken ages if I were to do it by myself (or if anyone had done this before).

It appears that

• \sum / \int (map-reduce of +) is the most used operator (\partial doesn't count because it appears 3 or 4 times in a line). This could have been matmul, but unfortunately this is implicit in latex.
• \alpha is actually being used more than \mu.
• \rangle happens more often than \psi (by a small margin)

[rht]

@jbenet https://api.github.com/users/rht/keys, the one with id 12491781, as @lgierth suggested.

[whyrusleeping]

@davidar the IPMR thing sounds like it could be accomplished with ethereum and ipfs.

[whyrusleeping]

@whyrusleeping what's the best way to proceed with adding lots of small files at the moment? Edit: use the tar adder? Are you able to do something like ipfs cat /ipfs/hash/archive.tar/foo/bar.txt?

The most reliable way is going to be to add the content offline. Even the tar add command will hit the file descriptor issue :/

We should get @rht access on pollux

Will take care of it this sprint: ipfs/infra#87

[jbenet]

it would be cool if you could pay nodes (bitcoin?) to run some sandboxed code over their local blockstore and publish the results back to the ipfs network. @jbenet have you had any thoughts on this topic?

that's like phase 3 of the IPPlan. (we're in phase 1). we could run some experiments now with this. it shouldn't be too hard.

[davidar]

The most reliable way is going to be to add the content offline. Even the tar add command will hit the file descriptor issue :/

@whyrusleeping I am, still too slow :(

Re ethereum, can you elaborate? I've played with it a little bit, and it seems incapable of doing any kind of interesting computation before heat death of the universe...

@jbenet cool! What's phase 2?

[davidar]

There's still a little bit of work that needs to be done, but I'm going to go ahead and call this the first complete (pdf+src+metadata) release of the arXiv archive:

https://ipfs.io/ipfs/QmfXH9XtP7xmoTH8WAp4HNSduqWMwLTH8B8TvbTkdgzNAa

🎈 🎆

[jbenet]

🌟 🌠 🎈 👍

[NDuma]

Link to GitHub via GitXiv.com

[davidar]

@NDuma GitXiv looks really cool, thanks! cc @samim23 @mekarpeles

[rht]

@davidar Request for reproducible build script. The spec for the package metadata (which includes the build process) should be achievable at 25 GB-scale (~ one server-node) regardless of the complication for many server-nodes case.

[davidar]

@rht I'm not sure there is a script (yet), it was mostly a (semi-)manual process... :/

[rht]

I see. I have re-created the scholarpedia.org archive in a local branch, since this is ~1GB, fits within one node of a computer. I wasn't able to find the example _Metadata.json from @eminence since it looks like the content has been gc-ed.

[leerspace]

The hash for this on archives.ipfs.io (/ipfs/QmfXH9XtP7xmoTH8WAp4HNSduqWMwLTH8B8TvbTkdgzNAa) seems to have become unavailable. If anyone still has a copy I can pin it to my node which I plan to keep running indefinitely.

[whyrusleeping]

I may have a copy on my NAS box which is currently in a storage unit (Recently moved). If nobody comes forth with another copy in the next couple months, i'll venture out to get it and plug it into a fat pipe

[razum2um]

Seems like QmfXH9XtP7xmoTH8WAp4HNSduqWMwLTH8B8TvbTkdgzNAa is still unavailable, isn't it?

[Stebalien]

I've managed to find a piece of it. I'm going to see if I can find any more.

[razum2um]

@Stebalien 18.91 MiB? yep, got it and nothing more for days