AMICO installation instruction for windows x64:

README.md

AMICO installation instruction for windows x64:

• conda create -n py27 python=2.7
• activate py27
• conda install -c intel mkl
• conda install numpy scipy six
• pip install nibabel
• conda install -c conda-forge dipy
• pip install https://github.com/samuelstjean/spams-python/releases/download/v2.6/spams-2.4-cp27-none-win_amd64.whl
• pip install git+https://github.com/daducci/AMICO.git
• pip install -U nibabel
• locate spams.py in the conda environment py27:
  • update line 424 in spams.py: from if D == None: to if D is None:

demo instruction:

• download example dataset https://github.com/daducci/AMICO/blob/master/doc/demos/NODDI_01.md
• move above data to c:\Temp and follow the folder structure from above link
• copy test_amico.py to c:\Temp
• make sure conda environment py27 is activated
• run python test_amico.py

I noticed that you had to edit some wheel files, by curiosity does it still works on recent numpy (say >1.12)? I remember they changed C/Fortran stuff ordering far in the backend and the whole thing had to be rebuilt with patches, but I forgot about the windows version as I don't have a compiler setup and all.

This was tested recently, so it should work with recent numpy, further, the code change to spams.py is a syntax error, so I do not believe it would interfere with change in numpy versions.

--
just checked the anaconda env used to test out these packages, numpy version is 1.13.1. :)
the conda list output is shown below.

--

packages in environment at d:\conda\envs\py27amico:

amico 1.0
backports 1.0 py27_1 conda-forge
backports.functools_lru_cache 1.4 py27_1 conda-forge
backports_abc 0.5 py27_0 conda-forge
bzip2 1.0.6 vc9_1 [vc9] conda-forge
ca-certificates 2017.11.5 0 conda-forge
certifi 2016.2.28 py27_0
cycler 0.10.0 py27_0 conda-forge
dipy 0.13.0 py27_vc9_1 [vc9] conda-forge
freetype 2.8.1 vc9_0 [vc9] conda-forge
functools32 3.2.3.2 py27_1 conda-forge
future 0.16.0 py27_0 conda-forge
h5py 2.7.1 py27_2 conda-forge
hdf5 1.10.1 vc9_1 [vc9] conda-forge
icu 58.2 vc9_0 [vc9] conda-forge
jpeg 9b vc9_2 [vc9] conda-forge
libpng 1.6.34 vc9_0 [vc9] conda-forge
matplotlib 2.1.2 py27_0 conda-forge
mkl 2017.0.3 0
nibabel 2.0.2 py27_vc9_0 [vc9] conda-forge
nibabel 2.2.1
numexpr 2.6.4 py27_0 conda-forge
numpy 1.13.1 py27_0
openssl 1.0.2n vc9_0 [vc9] conda-forge
pip 9.0.1 py27_1
pydicom 0.9.9 py27_0 conda-forge
pyparsing 2.2.0 py27_0 conda-forge
pyqt 5.6.0 py27_4 conda-forge
pytables 3.4.2 py27_7 conda-forge
python 2.7.13 1
python-dateutil 2.6.1 py27_0 conda-forge
pytz 2017.3 py_2 conda-forge
qt 5.6.2 vc9_2 [vc9] conda-forge
scikit-learn 0.19.0 np113py27_0
scipy 0.19.1 np113py27_0
setuptools 36.4.0 py27_1
singledispatch 3.4.0.3 py27_0 conda-forge
sip 4.18 py27_1 conda-forge
six 1.10.0 py27_0
spams 2.4
sqlite 3.13.0 vc9_1 [vc9] conda-forge
ssl_match_hostname 3.5.0.1 py27_1 conda-forge
tornado 4.5.3 py27_0 conda-forge
vc 9 0
vs2008_runtime 9.00.30729.5054 0
vtk 7.0.0 py27_vc9_1 [vc9] conda-forge
wheel 0.29.0 py27_0
wincertstore 0.2 py27_0
zlib 1.2.11 vc9_0 [vc9] conda-forge

I know they broke a bunch of stuff in numpy 1.10.x with the fortran ordering, but the old spams .so file is pretty much the same, I guess I can try hand editing the wheel file myself instead. Weird I had no comment reply notice or anything about this gist though.

There goes, I updated the wheel while with all the none checking changed, still passes all the tests on numpy 1.13.1.
It still the old version 2.4 and for python 2.7, but at least people won't need hand editing like in your last step from now on.