Lessons Learned: Installing SkFlow

SkFlow is the scikit-learn interface to Google’s opensource AI library TensorFlow. Installation started off straightforward enough and simply installing numpy and sklearn satisfied pre-requisites. Next we install TensorFlow with a simple:

sudo pip install –upgrade https://storage.googleapis.com/tensorflow/linux/cpu/tensorflow-0.7.1-cp27-none-linux_x86_64.whl

And last we install SkFlow itself with:

pip install skflow

Two stumbling blocks came up in by installation.

  • When installing Sklearn I got an error “no lapack/blas resources found“. The best solution I found was not re-installation, but simply installing the dev versions of the aforementioned libraries [1]:

    apt-get install libblas-dev liblapack-dev

  • After having installed Skflow, opening the python interpreter and issuing

    “import skflow”

    we get the following error:

    TypeError: <method ‘max’ of ‘numpy.ndarray’ objects> is not a Python function

    After tracing the library causing this error, the fix was to install “Dask” [2], the library at fault, by issuing:

    pip install dask[dataframe]

References

[1] https://groups.google.com/forum/#!topic/qiime-forum/8dVlTtlg59M

[2] http://dask.pydata.org/en/latest/install.html