Updated spaCy package on conda

[xpe]

What's the status of the conda spaCy channel? Who owns/maintains it? Is it "official"?

The last update (0.101.0) was 8 months ago according to https://anaconda.org/spacy/spacy/files.

It seems to me that pip is the only supported installation mechanism for the current releases of spaCy, since https://spacy.io/docs/usage/ says:

spaCy is compatible with 64-bit CPython 2.6+∕3.3+ and runs on Unix/Linux, macOS/OS X and Windows. The latest spaCy releases are currently only available as source packages over pip.

If I'm right, may I suggest a quick note/URL on the other locations (such as the conda spaCy channel) to help clarify?

[j-gyi]

It's be amazing if the conda spacy channel could be updated!

…

On Thu, Jan 26, 2017, 10:35 David James ***@***.***> wrote: What's the status of the conda spaCy channel <https://anaconda.org/spacy/spacy>? Who owns/maintains it? Is it "official"? The last update (0.101.0) was 8 months ago according to https://anaconda.org/spacy/spacy/files. It seems to me that pip is the only supported installation mechanism for the current releases of spaCy, since https://spacy.io/docs/usage/ says: spaCy is compatible with 64-bit CPython 2.6+∕3.3+ and runs on Unix/Linux, macOS/OS X and Windows. The latest spaCy releases are currently only available as source packages over pip. If I'm right, may I suggest a quick note/URL on the other locations (such as the conda spaCy channel) to help clarify? — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#778>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AQ9785b46M4XMxvsO7Tr_m3OOBrBqoEWks5rWCmkgaJpZM4LuTSo> .

[honnibal]

Hey,

I've wanted to set up properly automatic builds again for spaCy, which would include builds for conda. This never made it to the top of the priority queue unfortunately.

Including all the current requests I'd eventually like to satisfy, a full build matrix has the following:

Channels

• pip
• conda
• docker
• ami
• deb

Python versons

• 2.7
• 3.5
• 3.6

Operating system

• Linux
• Windows
• OSX

Compiler

• gcc
• clang
• Visual Studio
• cygwin

Architecture

• 32 bit Python
• 64 bit Python

Total configurations

5 * 3 * 3 * 4 * 2 = 360 build configurations.

Docker, AMI and deb are requested less than conda, so assume these are the lowest priorities. That still gives 144 things to test in the full build matrix, though. This is why Windows support has been a bit shaky, as has support for 32bit Pythons.

Possible solutions

In the absence of a systematic build service, we could still manually push each version to conda, the way we do to pip. It's not fantastic to add extra manual steps to the release process though, and we also maintain separate packages for spaCy's dependencies, specifically the thinc, preshed and cymem libraries. preshed and cymem aren't updated often, but thinc is currently under very active development.

The other solution is to only publish through pip, and ask conda users to do pip install. I must confess that I don't actually use conda myself. How does being able to write conda install spacy instead of pip install spacy improve things for you?

I also want to clarify that unfortunately this is something that inherently has to be centralised...The community can't easily help here, because all of the binaries and source packages have to be built by Explosion.

[j-gyi]

Hey just that I often have trouble with pip vs conda - I'm new to coding and like to keep things as simple as possible. - though spacy doesn't have conflicting packages generally speaking pip played a huge part in my growing pains in understanding the importance of separate environments. not using pip just gives me assurance that no other package has been affected - my main purpose of coding is for data analytics so anaconda pretty much gives me everything i need. - When i use conda install things usually work out smoothly, where as when I pip install, its not as consistent - with spacy the pip install worked when i tried on ubuntu, but not on debian, which i am using, it repeatedly gave an error when running the `python -m spacy.en.download all`, command. But i'm pretty sure if I had run in with a conda command everything would have went more smoothly. though I don't have the exact error message, i removed everything to start fresh before I realized I'd be writing this

…

On Thu, Jan 26, 2017 at 3:11 PM Matthew Honnibal ***@***.***> wrote: Hey, I've wanted to set up properly automatic builds again for spaCy, which would include builds for conda. This never made it to the top of the priority queue unfortunately. Including all the current requests I'd eventually like to satisfy, a full build matrix has the following: Channels - pip - conda - docker - ami - deb Python versons - 2.7 - 3.5 - 3.6 Operating system - Linux - Windows - OSX Compiler - gcc - clang - Visual Studio - cygwin Architecture - 32 bit Python - 64 bit Python Total configurations 5 * 3 * 3 * 4 * 2 = *360* build configurations. Docker, AMI and deb are requested less than conda, so assume these are the lowest priorities. That still gives 144 things to test in the full build matrix, though. This is why Windows support has been a bit shaky, as has support for 32bit Pythons. Possible solutions In the absence of a systematic build service, we could still manually push each version to conda, the way we do to pip. It's not fantastic to add extra manual steps to the release process though, and we also maintain separate packages for spaCy's dependencies, specifically the thinc, preshed and cymem libraries. preshed and cymem aren't updated often, but thinc is currently under very active development. The other solution is to only publish through pip, and ask conda users to do pip install. I must confess that I don't actually use conda myself. How does being able to write conda install spacy instead of pip install spacy improve things for you? I also want to clarify that unfortunately this is something that inherently has to be centralised...The community can't easily help here, because all of the binaries and source packages have to be built by Explosion. — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#778 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AQ9784dWrKmTs3tX0QOf2depSw70-rN_ks5rWGoygaJpZM4LuTSo> .

[j-gyi]

but yes that's a low of work to get the conda channel going! I had no idea On Thu, Jan 26, 2017 at 5:01 PM Douglas Owusu <douglas@douglasowusu.com> wrote:

…

Hey just that I often have trouble with pip vs conda - I'm new to coding and like to keep things as simple as possible. - though spacy doesn't have conflicting packages generally speaking pip played a huge part in my growing pains in understanding the importance of separate environments. not using pip just gives me assurance that no other package has been affected - my main purpose of coding is for data analytics so anaconda pretty much gives me everything i need. - When i use conda install things usually work out smoothly, where as when I pip install, its not as consistent - with spacy the pip install worked when i tried on ubuntu, but not on debian, which i am using, it repeatedly gave an error when running the `python -m spacy.en.download all`, command. But i'm pretty sure if I had run in with a conda command everything would have went more smoothly. though I don't have the exact error message, i removed everything to start fresh before I realized I'd be writing this On Thu, Jan 26, 2017 at 3:11 PM Matthew Honnibal ***@***.***> wrote: Hey, I've wanted to set up properly automatic builds again for spaCy, which would include builds for conda. This never made it to the top of the priority queue unfortunately. Including all the current requests I'd eventually like to satisfy, a full build matrix has the following: Channels - pip - conda - docker - ami - deb Python versons - 2.7 - 3.5 - 3.6 Operating system - Linux - Windows - OSX Compiler - gcc - clang - Visual Studio - cygwin Architecture - 32 bit Python - 64 bit Python Total configurations 5 * 3 * 3 * 4 * 2 = *360* build configurations. Docker, AMI and deb are requested less than conda, so assume these are the lowest priorities. That still gives 144 things to test in the full build matrix, though. This is why Windows support has been a bit shaky, as has support for 32bit Pythons. Possible solutions In the absence of a systematic build service, we could still manually push each version to conda, the way we do to pip. It's not fantastic to add extra manual steps to the release process though, and we also maintain separate packages for spaCy's dependencies, specifically the thinc, preshed and cymem libraries. preshed and cymem aren't updated often, but thinc is currently under very active development. The other solution is to only publish through pip, and ask conda users to do pip install. I must confess that I don't actually use conda myself. How does being able to write conda install spacy instead of pip install spacy improve things for you? I also want to clarify that unfortunately this is something that inherently has to be centralised...The community can't easily help here, because all of the binaries and source packages have to be built by Explosion. — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#778 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AQ9784dWrKmTs3tX0QOf2depSw70-rN_ks5rWGoygaJpZM4LuTSo> .

[honnibal]

To be clear: It's not that there's so much work to get the conda channel going specifically. It's just that there are lots of different dimensions to the build matrix, and we want to do less if we can.

I can't say exactly what your problem might be with the pip install. I usually use virtualenv to keep my projects separate.

[xpe]

@honnibal -- thanks for explaining the moving parts here and the relative importance of each!

For those who haven't compared conda, pip, and virtualenv, may I recommend reading the short summary at conda vs. pip vs. virtualenv? It has a nice table showing the commands for each tool (pip, virtualenv, conda). It says it quite well:

If you've used pip and virtualenv in the past, you can use conda to perform all of the same operations. Pip is a package manager, and Virtualenv is an environment manager. Conda is both.

Conda was designed to overcome many of the challenges of pip and virtualenv. I use it and recommend it. I'd need to review the specifics to fully explain the advantages. Happy to if that helps.

[elyase]

I normally prefer to use conda packages but lately most scientific packages I use also work well with pip. spacy in particular has always worked well with pip for me.

[rmax]

A conda package for spacy (and textacy) will be added soon to the conda-forge channel: conda-forge/staged-recipes#2448

Newer releases will be published afterwards. Feel free to ping me if you want to be added as recipe maintainer.

[ines]

@rolando Thanks a lot for taking this on – much appreciated! Once it's set up, we'd be happy to help maintain it. It seems like this will hopefully make things much easier.

Btw, I've already set up a new organisation on Anaconda Cloud which we can use for distribution in the future. (Unfortunately, we no longer have access to the old spaCy channel as the account that created it no longer exists. I'll send a support email, though – maybe there's a way to merge the two accounts.)

[rmax]

Packages are up and the most installed package so far has been for win-64 platform: https://anaconda.org/conda-forge/spacy

Install command: conda install -c conda-forge spacy
Feedstock: https://github.com/conda-forge/spacy-feedstock

Newer versions are on the way.

[ines]

@rolando Thanks a lot, this is great! Just updated the README to add installation instructions (see 33efe77). Is there anything else we should mention that's relevant to conda users?

[rmax]

@ines the installation instructions look good! It already mentions the repository for recipe contributions so that should be enough for conda users.

[ines]

Perfect! I'll update the website & close this issue! 🎉

[lock]

This thread has been automatically locked since there has not been any recent activity after it was closed. Please open a new issue for related bugs.