Missing scikit-image on Windows

James_Chang · ‎09-15-2017

Both release note of IDP3 2017 update3 and 2018 state that scikit-image 0.13.0 is available, but actually it is not. After IDP3 installed, I have to do conda install to get scikit-image with some packages downgraded:

The following NEW packages will be INSTALLED:

    icu:          57.1-vc14_0                   [vc14      ]
    jpeg:         9b-vc14_0                     [vc14      ]
    libtiff:      4.0.6-vc14_3                  [vc14      ]
    networkx:     1.11-py36_0
    olefile:      0.44-py36_0
    pillow:       4.2.1-py36_0
    pyqt:         5.6.0-py36_2
    pywavelets:   0.5.2-np113py36_0
    qt:           5.6.2-vc14_6                  [vc14      ]
    scikit-image: 0.13.0-np113py36_0
    sip:          4.18-py36_0

The following packages will be UPDATED:

    conda:        4.3.16-py36_intel_1     intel [intel     ] --> 4.3.25-py36_0

The following packages will be SUPERSEDED by a higher-priority channel:

    conda-env:    2.6.0-0                 intel --> 2.6.0-0

The following packages will be DOWNGRADED due to dependency conflicts:

    freetype:     2.8-vc14_intel_0        intel [vc14 intel] --> 2.5.5-vc14_2               [vc14      ]
    hdf5:         1.8.17-intel_9          intel [intel     ] --> 1.8.17-vc14_intel_8  intel [vc14 intel]
    libpng:       1.6.30-intel_0          intel [intel     ] --> 1.6.28-vc14_intel_10 intel [vc14 intel]
    matplotlib:   2.0.2-np113py36_intel_1 intel [intel     ] --> 2.0.2-np113py36_0
    python:       3.6.2-intel_3           intel [intel     ] --> 3.6.2-0
    tk:           8.6.4-vc14_intel_26     intel [vc14 intel] --> 8.5.18-vc14_0              [vc14      ]

---------------------------------------------------------------------------------------------------

I tried to create a new environment from Anaconda 4.4 with "conda create ‐n idp intelpython3_full python=3", but it got an UnsatisfiableError:

UnsatisfiableError: The following specifications were found to be in conflict:
  - intelpython3_core
  - python 3.6*
  - scikit-image
Use "conda info <package>" to see the dependencies for each package.

How could I get the official scikit-image package installed? Thanks.

Todd_T_Intel · ‎09-15-2017

Chin Yuan,

Thank you for using the Intel(R) Distribution for Python.

There are a couple things leading to your problem. The first is that our Release Notes are mistaken -- we do not have scikit-image packages for Windows or macOS, only for Linux. I will make sure that is corrected.

The reason your attempt to install scikit-image from the defaults channel failed is a little more complex. Scikit-image has multiple package requirements that we do not provide, and some of them have very specific version requirements that conflict with versions specified in the intelpython_full metapackage. For example, scikit-image requires pillow, and pillow requires freetype 2.5.*. We ship freetype 2.8 and our matplotlib package requires freetype 2.8. That is a conflict that conda needs to try to resolve.

The reason conda could not find a good resolution is the intelpython3_full and intelpython3_core packages require specific versions of packages and did not have a version that would meet the installation request requirements. That problem has been resolved with the addition of a version of the intelpython3_core package called "0.0.0.custom". Now if you attempt to install a package that conflicts with the specific version of intelpython3 in your environment, conda can fallback to the 0.0.0.custom versions and satisfy your request.

Now, issuing a "conda install scikit-image -c intel" command will succeed, with multiple new package installations and some version changes. The changes are unavoidable due to very specific dependencies, but I saw no changes that should cause you problems.

(idp3) C:\localdisk\pybuilds>conda install scikit-image
Fetching package metadata ...............
Solving package specifications: .

Package plan for installation in environment C:\localdisk\work\Miniconda3\envs\idp3:

The following NEW packages will be INSTALLED:

    icu:               57.1-vc14_0                   [vc14      ]
    jpeg:              9b-vc14_0                     [vc14      ]
    libtiff:           4.0.6-vc14_3                  [vc14      ]
    menuinst:          1.4.7-py36_0
    networkx:          1.11-py36_0
    olefile:           0.44-py36_0
    pillow:            4.2.1-py36_0
    pycosat:           0.6.2-py36_0
    pyqt:              5.6.0-py36_2
    pywavelets:        0.5.2-np113py36_0
    pywin32:           220-py36_2
    qt:                5.6.2-vc14_6                  [vc14      ]
    scikit-image:      0.13.0-np113py36_0
    sip:               4.18-py36_0

The following packages will be DOWNGRADED:

    freetype:          2.8-vc14_intel_0        intel [vc14 intel] --> 2.5.5-vc14_2               [vc14      ]
    hdf5:              1.8.17-intel_9          intel [intel     ] --> 1.8.17-vc14_intel_8  intel [vc14 intel]
    intelpython3_core: 2018.0.0-0              intel --> 0.0.0.custom-0       intel
    intelpython3_full: 2018.0.0-0              intel --> 0.0.0.custom-0       intel
    libpng:            1.6.30-intel_0          intel [intel     ] --> 1.6.28-vc14_intel_10 intel [vc14 intel]
    matplotlib:        2.0.2-np113py36_intel_1 intel [intel     ] --> 2.0.2-np113py36_0
    python:            3.6.2-intel_3           intel [intel     ] --> 3.6.2-0
    tk:                8.6.4-vc14_intel_26     intel [vc14 intel] --> 8.5.18-vc14_0              [vc14      ]

Proceed (/n)? y

James_Chang · ‎09-15-2017

Thank you Todd,

I got scikit-image installed into IDP3, just thought that there maybe a secret method to install the official intel packages without downgrading some of them.

For Anaconda 4.4, I can also install scikit-image into the intel channel environment after I do "conda update conda" from root environment, and of course also with some packages downgraded.