Hi there, just saw your Q via the mailing list. So, the Travis issue is due to pep8 violations (a Python style guide, more info here: https://www.python.org/dev/peps/pep-0008/)

E.g., if you look at the logs in the travis report, you see sth like

> examples/cluster/plot_coin_ward_segmentation.py:39:63: E231 missing whitespace after ','
rescaled_coins = rescale(smoothened_coins, 0.2, mode="reflect",multichannel=False,anti_aliasing=False)
                                                              ^
examples/cluster/plot_coin_ward_segmentation.py:39:80: E501 line too long (102 > 79 characters)
rescaled_coins = rescale(smoothened_coins, 0.2, mode="reflect",multichannel=False,anti_aliasing=False)
                                                                               ^
examples/cluster/plot_coin_ward_segmentation.py:39:82: E231 missing whitespace after ','
rescaled_coins = rescale(smoothened_coins, 0.2, mode="reflect",multichannel=False,anti_aliasing=False)
                                                                                ^

That essentially indicates you are missing a white space between the function arguments. I.e., instead of

mode="reflect",multichannel=False,anti_aliasing=False

write

mode="reflect", multichannel=False, anti_aliasing=False

I recommend using a pep8 or flake8 plugin for your text editor that highlights these issues. Or you could run pep8 or flake8 tool to before resubmitting the code to make sure you catch all of these. Flake8 is basically an advanced version of the pep8 tool (= a tool for Python that checks for PEP8 style), where it also checks for unused variable names, unused imports, etc.

After installing flake8 (https://pypi.org/project/flake8/; e.g., via pip install flake8), you should be able to call it as a command line tool, e.g., flake8 sklearn/.../your-code.py

Thank you very much @rasbt ! That makes so much sense now !

No problem. The reason for the other error (the unrecognized multichannel arg in skimage's rescale) is prob because scikit-image used in the CircleCI build is simply too old. This would need to be updated here:

but should probably be discussed with the core devs

@amueller @jnothman is it possible to update the version of scikit-image used in CircleCi ?

sure. check build_tools/circle
@jnothman This is the only piece of code I see in build_tools/circle that is related to scikit-image version.
conda create -n $CONDA_ENV_NAME --yes --quiet python="${PYTHON_VERSION:-*}" \ numpy="${NUMPY_VERSION:-*}" scipy="${SCIPY_VERSION:-*}" cython \ pytest coverage matplotlib="${MATPLOTLIB_VERSION:-*}" sphinx=1.6.2 pillow \ scikit-image="${SCIKIT_IMAGE_VERSION:-*}" pandas="${PANDAS_VERSION:-*}"

And I think the value for ${SCIKIT_IMAGE_VERSION:-*} is coming from the file .circleci/congif.yml (SCIKIT_IMAGE_VERSION: 0.9.3 - line 49)
Since this is supposed to be a hidden file , I'm not sure if I have the permission to make any changes.

I actually don't think we can nicely resolve this yet

Agreed. IIRC our dependencies versions are based on what the packages version is in Ubuntu 14.04 hence scikit-image 0.9.3.

More generally, getting rid of deprecation warnings and keeping compatibility with old versions are going to be at odds with each other. You could potentially complicate the example code (e.g. by adding an if clause based on the version of scikit-image in this case) but complicating an example without a very good reason is something we try to avoid.

This was resolved in #12654 . Closing. Thanks for contributing.