I do not have a strong opinion on this. Since the anaconda-client package is python-only, installing from git seems okay with me.

In the travis case, downloading and install miniforge would add a little bit of time to the build and I think we want to conserve our credits there.

I tried to install anaconda-project before anaconda-client from git (as requested by the requirements.txt file) and I ran into the error shown here: https://github.com/scikit-learn/scikit-learn/runs/1959589120.

Nevertheless, I ran:

pip install git+https://github.com/Anaconda-Platform/anaconda-project
pip install git+https://github.com/Anaconda-Platform/anaconda-client

in my machine and it works.

Not sure what is the problem in Github Actions.

Ah I see, if installing with anaconda-client from conda fixes a bug, then I am +1 with this PR.

Actually, I am confused, none of the above [cd build] ran successfully and the one with conda caused:

 anaconda: error: argument : invalid choice: 'scikit-learn-wheels-staging' (choose from 'auth', 'label', 'channel', 'config', 'copy', 'download', 'groups', 'login', 'logout', 'move', 'notebook', 'package', 'remove', 'search', 'show', 'upload', 'whoami')
Error: Process completed with exit code 2.

so maybe the the command-line parsing has changed. You can test it locally by creating your own personal anaconda.org channel.

Sorry for the late reply. The actual problem is that ANACONDA_TOKEN is empty (secret) so the command interprets upload as token for the -t option .

Let me check for a personal project.

What is the status of this @alfaro96? Does it works on the CI configured on your repo?

Let me run the Wheel builder workflow with the latest commit and I will let you know the result ASAP.

Here is the workflow: https://github.com/alfaro96/scikit-learn/runs/2115411152?check_suite_focus=true.

It is failing, but I think that it is related with the token.

Edit: I have figured out that the error was related with a bug in Python 3.9. I have downgraded to Python 3.8 and it is working. Here is the working run: https://github.com/alfaro96/scikit-learn/actions/runs/657229124.

Edit: I have figured out that the error was related with a bug in Python 3.9. I have downgraded to Python 3.8 and it is working. Here is the working run: https://github.com/alfaro96/scikit-learn/actions/runs/657133352.

This is actually failing:

build_tools/github/upload_anaconda.sh: line 11: syntax error near unexpected token `fi'
Error: Process completed with exit code 2.

Edit: I have figured out that the error was related with a bug in Python 3.9. I have downgraded to Python 3.8 and it is working. Here is the working run: https://github.com/alfaro96/scikit-learn/actions/runs/657133352.

This is actually failing:

build_tools/github/upload_anaconda.sh: line 11: syntax error near unexpected token `fi'
Error: Process completed with exit code 2.

Yep, I have committed a new version that I hope does not fail 😅.

Let us wait for the https://github.com/alfaro96/scikit-learn/actions/runs/657229124 to finish and, if it works, I will clean the files changed 😉.

Okay, the workflow (https://github.com/alfaro96/scikit-learn/actions/runs/657229124) is working and the wheels properly uploaded to the corresponding index: https://pypi.anaconda.org/alfaro96/simple/scikit-learn/.

Let me clean the files changed for preparing to merge.

Thank you very much @alfaro96.