Reproducible machine learning with PyTorch and Quilt

#1

Post by akarve » Tue, Jul 17, 2018, 5:22 PM UTC

Reproducible machine learning with PyTorch and Quilt
blog.paperspace.com

Re: Reproducible machine learning with PyTorch and Quilt

#2

Post by cwyers » Tue, Jul 17, 2018, 5:54 PM UTC

It seems to me like the machine learning algorithm here is mostly learning how to add JPEG compression artifacts to images.

Re: Reproducible machine learning with PyTorch and Quilt

#3

Post by p1esk » Tue, Jul 17, 2018, 5:55 PM UTC

Oh, this resonates with me so much! I'm running 4 different DeepSpeech models right now, each using a differently processed version of LibriSpeech dataset (mfcc/fbanks/linear spectrograms, deltas? energy? padding? etc). Because the original DS papers didn't bother describing it, and every implementation I found uses completely different methods and libraries.

Not to mention every one of those implementation packages their preprocessed version into a different data format, and then creates a different data pipeline (and I only looked at tensorflow implementations)

Re: Reproducible machine learning with PyTorch and Quilt

#4

Post by rhacker » Tue, Jul 17, 2018, 6:48 PM UTC

Please please please don't kill our favorite plot device. Make sure the process takes exactly 3 days.

Re: Reproducible machine learning with PyTorch and Quilt

#5

Post by jononor » Tue, Jul 17, 2018, 6:54 PM UTC

Was not aware of Quilt for hosting datasets. Is it the go-to in this area? What other alternatives are there?

Re: Reproducible machine learning with PyTorch and Quilt

#6

Post by eindiran » Tue, Jul 17, 2018, 7:53 PM UTC

jononor wrote: post #5

Was not aware of Quilt for hosting datasets. Is it the go-to in this area? What other alternatives are there?

You can use AWS to host open datasets: https://aws.amazon.com/opendata/public-datasets/

These are some other people working in roughly the same space: http://datproject.org/ http://www.pachyderm.io/

But it does seem like Quilt is a go-to, if you are looking for a "Github for data" host.

Re: Reproducible machine learning with PyTorch and Quilt

#7

Post by infinity0 » Tue, Jul 17, 2018, 8:15 PM UTC

A step in the right direction for machine learning in science, but they could have done some more research into naming conflicts:

$ apt-cache show quilt

Package: quilt

[..]

Description-en: Tool to work with series of patches

Quilt manages a series of patches by keeping track of the changes each of them makes. They are logically organized as a stack, and you can apply, un-apply, refresh them easily by traveling into the stack (push/pop). . Quilt is good for managing additional patches applied to a package received as a tarball or maintained in another version control system. The stacked organization is proven to be efficient for the management of very large patch sets (more than hundred patches). As matter of fact, it was designed by and for Linux kernel hackers (Andrew Morton, from the -mm branch, is the original author), and its main use by the current upstream maintainer is to manage the (hundreds of) patches against the kernel made for the SUSE distribution. . This package provides seamless integration into Debhelper or CDBS, allowing maintainers to easily add a quilt-based patch management system in their packages. The package also provides some basic support for those not using those tools. See README.Debian for more information.

$ zcat /usr/share/doc/quilt/changelog.gz | tail -n3

Version 0.26 (Tue Oct 21 2003) - Change summary not available

Re: Reproducible machine learning with PyTorch and Quilt

#8

Post by jmaxfield » Tue, Jul 17, 2018, 9:06 PM UTC

eindiran wrote: post #6

jononor wrote: post #5

Was not aware of Quilt for hosting datasets. Is it the go-to in this area? What other alternatives are there?

You can use AWS to host open datasets: https://aws.amazon.com/opendata/public-datasets/ These are some other people working in roughly the same space: http://datproject.org/ http://www.pachyderm.io/ But it does seem like Quilt is a go-to, if you are looking for a "Github for data" host.

I use Quilt pretty much daily and while I like AWS open datasets I don't think it is as actively developed on as Quilt is. DAT project on the other hand I really do like as a way to simply transfer large amounts of data between contributors, that said, if you are just trying to get data out there and have people use it freely for their own work I think Quilt presents the solution due to searchable and easily understood python (and I think an R repo) usage of datasets.

Re: Reproducible machine learning with PyTorch and Quilt

#9

Post by shoyer » Tue, Jul 17, 2018, 9:24 PM UTC

jononor wrote: post #5

Was not aware of Quilt for hosting datasets. Is it the go-to in this area? What other alternatives are there?

A couple of publicly available alternatives I'm aware of include:

- Girder: http://girder.readthedocs.io

- Intake: https://github.com/ContinuumIO/intake

I haven't used any of these, but I agree that the idea is quite compelling.

Re: Reproducible machine learning with PyTorch and Quilt

#10

Post by jdoliner » Tue, Jul 17, 2018, 9:25 PM UTC

eindiran wrote: post #6

jononor wrote: post #5

Was not aware of Quilt for hosting datasets. Is it the go-to in this area? What other alternatives are there?

You can use AWS to host open datasets: https://aws.amazon.com/opendata/public-datasets/ These are some other people working in roughly the same space: http://datproject.org/ http://www.pachyderm.io/ But it does seem like Quilt is a go-to, if you are looking for a "Github for data" host.

Pachyderm founder here. We're not really a data hosting provider, although we may offer that in the future. Right now Pachyderm is more like intranet data hosting for companies. You have to spin up your own Kubernetes cluster and deploy Pachyderm on it. It's also not normally used to download data onto your local machine for processing because it has its own computation layer which allows you to run code at scale and tracks the provenance of the data to keep things reproducible.