[MRG + 1] BLD: Add 'pip' extra for installing with pip #6990
Conversation
|
LGTM if you tested it ;) |
|
Yup. If I do |
|
What do you want documentation-wise? I can update https://github.com/scikit-learn/scikit-learn/blob/master/doc/install.rst, but when I ran into this problem originally, I don't think I would have even bothered to look there – I'd have much more likely looked at the |
|
can confirm that this works on my machine, i just tried it out. thanks for the fix |
|
One more possible nit – this could be structured as e.g. |
|
@jnothman wdyt? |
|
Following up here – this would let us clean up our install procedures a bit. |
|
I like moving the dependency versions to the top of the file. I think we are increasingly able to rely on pip, and this is a decent transition. Yes, needs documentation for it to be useful. |
|
Does |
|
Or |
|
|
|
Updated. I added a note to |
|
still LGTM. |
| your operation system, it is recommended to use a third party distribution. | ||
| your operating system, we recommended using a third-party distribution. | ||
|
|
||
| If you must install scikit-learn with pip, you can install it as |
There was a problem hiding this comment.
I'll go with "and its dependencies". The target use case isn't typing pip install scikit-learn[alldeps] on the command line – it's instead having a single requirements.txt that lists everything.
There was a problem hiding this comment.
Clarify that then?
On 23 August 2016 at 12:55, Jimmy Jia notifications@github.com wrote:
In doc/install.rst
#6990 (comment)
:If you are using Linux, consider using your package manager to install
scikit-learn. It is usually the easiest way, but might not provide the newest
version.
If you haven't already installed numpy and scipy and can't install them via
-your operation system, it is recommended to use a third party distribution.
+your operating system, we recommended using a third-party distribution.
+
+If you must install scikit-learn with pip, you can install it asI'll go with "and its dependencies". The target use case isn't typing pip
install scikit-learn[alldeps] on the command line – it's instead having a
single requirements.txt that lists everything.—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub
https://github.com/scikit-learn/scikit-learn/pull/6990/files/83d14b234d3be76737a15f12f8dc99696d0bf4d3#r75795830,
or mute the thread
https://github.com/notifications/unsubscribe-auth/AAEz64iSp68aVAfdWDAGSObo9_PV2MIzks5qimEtgaJpZM4JM0Pv
.
There was a problem hiding this comment.
Updated. This now reads:
If you must install scikit-learn and its dependencies with pip, you can install
it asscikit-learn[alldeps]. We strongly recommend against doing this
unless you are familiar with how to correctly build numpy and scipy. The most
common use case for this is in arequirements.txtfile used as part of an
automated build process for a PaaS application or a Docker image. This option
is not intended for manual installation from the command line.
This allows advanced users who are using standard Python package management tools to install scikit-learn[alldeps], which includes the appropriate numpy and scipy dependencies.
|
I think it would be good to have that for the release. @jnothman ? |
|
LGTM |
|
thanks @taion |
This allows advanced users who are using standard Python package management tools to install scikit-learn[alldeps], which includes the appropriate numpy and scipy dependencies.
This allows advanced users who are using standard Python package management tools to install scikit-learn[alldeps], which includes the appropriate numpy and scipy dependencies.
Reference Issue
Fixes #4164
What does this implement/fix? Explain your changes.
This allows advanced users who are using standard Python management tools to install
scikit-learn[pip], which includes the appropriatenumpyandscipydependencies.