choderalab / ensembler

Automated omics-scale protein modeling and simulation setup.
http://ensembler.readthedocs.io/
GNU General Public License v2.0
52 stars 21 forks source link

'module' object has no attribute 'Ward' #59

Closed rafwiewiora closed 8 years ago

rafwiewiora commented 8 years ago

Just want to mention that currently conda installation is not functional due to the cluster.Ward deprecation in scikit-learn 0.17 and hence the exception thrown by msmbuilder.

See here: https://github.com/msmbuilder/msmbuilder/issues/692

Installation of msmbuilder from Github repo rescues it.

jchodera commented 8 years ago

Can you try again? Remove your github installation of msmbuilder (will have to do this manually) and try

conda uninstall ensembler msmbuilder mdtraj scikit-learn --yes
conda clean -plts --yes
conda install ensembler --yes

They have fixed the conda package for msmbuilder so that it should work now without changes.

rafwiewiora commented 8 years ago

Not quite, the msmbuilder is not fixed in the conda build yet, all that seems to have been done is forcing scikit-learn 0.16 rather than 0.17 as they mention in msmbuilder/msmbuilder#692.

But:

conda install ensembler --yes

installs scikit-learn 0.17 as dependency.

Still have to do:

conda remove scikit-learn --yes
conda install scikit-learn=0.16
jchodera commented 8 years ago

Not quite, the msmbuilder is not fixed in the conda build yet, all that seems to have been done is forcing scikit-learn 0.16 rather than 0.17 as they mention in msmbuilder/msmbuilder#692.

Any idea why this is? ensembler doesn't specify scikit-learn as a dependency, and msmbuilder specifies scikit-learn <=0.16, so unless you have an old cached package somewhere, it should be either pulling in 0.16 or earlier or say that it couldn't resolve a conflict in dependencies (if some other package is demanding 0.16).

Ah! I forgot the -i argument to conda clean:

conda uninstall ensembler msmbuilder mdtraj scikit-learn --yes
conda clean -pltis --yes
conda install ensembler --yes
rafwiewiora commented 8 years ago

and msmbuilder specifies scikit-learn <=0.16

This is the problem. conda install msmbuilder install the 3.3.0-np110py27_1 build, but they have corrected the scikit-learn dependency in the 3.3.0-np110py27_2 build - so conda install ensembler installs the _1 build and hence scikit-learn 0.17.

This is why I've asked this on Slack:

rafwiewiora [3:48 PM] What’s the conda command for installing a particular build?

conda install software-x=version works ok, but not for the situation with msmbuilder where they now have both msmbuilder-3.3.0-np110py27_1 and msmbuilder-3.3.0-np110py27_2 (conda install msmbuilder installs _1 build, conda install msmbuilder=3.3.0-np110py27_2 doesn’t work).

All I worked out for now is downloading the tar.bz2 from https://conda.anaconda.org/omnia/osx-64 and conda install msmbuilder-3.3.0-np110py27_2.tar.bz2

rafwiewiora commented 8 years ago
The following packages will be downloaded:

    package                    |            build
    ---------------------------|-----------------
    conda-3.18.6               |           py27_0         176 KB
    scikit-learn-0.17          |      np110py27_1         3.7 MB
    mdtraj-1.5.1               |      np110py27_0        16.6 MB
    msmbuilder-3.3.0           |      np110py27_1         1.0 MB
    ensembler-1.0.5            |           py27_0        23.9 MB
    ------------------------------------------------------------
                                           Total:        45.3 MB

The following NEW packages will be INSTALLED:

    ensembler:    1.0.5-py27_0     
    mdtraj:       1.5.1-np110py27_0
    msmbuilder:   3.3.0-np110py27_1
    scikit-learn: 0.17-np110py27_1 

The following packages will be UPDATED:

    conda:        3.18.5-py27_0 --> 3.18.6-py27_0    
jchodera commented 8 years ago

Can you post this in the msmbuilder issue tracker? My understanding is that conda is supposed to install the latest build, not pick whichever build it likes!

jchodera commented 8 years ago

We may need to have them delete the earlier build from the online anaconda.org space.

rafwiewiora commented 8 years ago

Yeah, that's what I thought too! Ok posting now.

rafwiewiora commented 8 years ago

This is good to close.