Build mlpy with python3.3 (64bit) (windows 64bit)

Build mlpy in windows 64bit environment

The binary for windows of mlpy is provided only for 32bit, so build it by yourself. usage environment:

• python3.3.5(64bit)
• windows7 64bit sp1
• Visual Studio 2010

Preparation

Get the library you need

• numpy
• scipy
• GSL

Install numpy, scipy

If you're trying to use mlpy, you probably already have it installed. If not, install it with pip install numpy scipy

GSL build

You will need a static library. I used to build another GSL more than half a year ago, but now it seems that things are changing.

Before

When I did it, I referred to the link below. It was surprisingly easy.

• Note on using GSL 1.15 with visual C ++
• Procedure for using GSL 1.16 with Visual Studio 2012

Current

Now the link to the visual studio project is gone. The version is a little old, but gsl-1.15-vc10.zip I found a page that distributes. There was also something that looked like an All in One package (gnu-gsl-for-windows). I'm not using either, so I'm not sure if it's really available.

• 2014/7/3 postscript: gnu-gsl-for-windows was usable as expected.

Fix mlpy

Modify the source code for 64-bit environment.

Changed " win32 " in setup.py to " win-amd64 ". However, the 25th line remains as it is

before

setup.py(29-41 lines)

#### libs
if get_platform() == "win32":
   gsl_lib = ['gsl', 'cblas']
   math_lib = []
else:
   gsl_lib = ['gsl', 'gslcblas']
   math_lib = ['m']
   
#### Extra compile args
if get_platform() == "win32":
   extra_compile_args = []
else:
   extra_compile_args = ['-Wno-strict-prototypes']

after

setup.py(29-41 lines)

#### libs
if get_platform() == "win-amd64":
   gsl_lib = ['gsl', 'cblas']
   math_lib = []
else:
   gsl_lib = ['gsl', 'gslcblas']
   math_lib = ['m']
   
#### Extra compile args
if get_platform() == "win-amd64":
   extra_compile_args = []
else:
   extra_compile_args = ['-Wno-strict-prototypes']

Fixed double-byte characters in comments. Removed Θ and double quotes (”)

before

mlpy/fastcluster/fastcluster/src/fastcluster.cpp(630-632 lines)

    // Complexity: Θ(size)
    // Reference: Cormen, Leiserson, Rivest, Stein, Introduction to Algorithms,
    // 3rd ed., 2009, Section 6.3 “Building a heap”

after Delete comment

Change np.int_t to np.int64_t

If you do not correct it, a runtime error will occur due to a mismatch between int type and long long type, and even Tutorial cannot be executed. __ There may be other areas that need to be fixed, but I haven't found them yet __ before

mlpy/libsvm/libsvm.pyx(28-32 lines)

# array 1D to svm node
cdef svm_node *array1d_to_node(np.ndarray[np.float64_t, ndim=1] x):
    cdef int i, k
    cdef np.ndarray[np.int_t, ndim=1] nz
    cdef svm_node *ret

after

mlpy/libsvm/libsvm.pyx(28-32 lines)

# array 1D to svm node
cdef svm_node *array1d_to_node(np.ndarray[np.float64_t, ndim=1] x):
    cdef int i, k
    cdef np.ndarray[np.int64_t, ndim=1] nz
    cdef svm_node *ret

before

mlpy/adatron/adatron.pyx(69-73 lines)

        
        cdef np.ndarray[np.int_t, ndim=1] ynew
        cdef np.ndarray[np.float_t, ndim=2] K_arr
        cdef np.ndarray[np.float_t, ndim=1] alpha_arr
        cdef double margin

after

mlpy/adatron/adatron.pyx(69-73 lines)

        
        cdef np.ndarray[np.int64_t, ndim=1] ynew
        cdef np.ndarray[np.float_t, ndim=2] K_arr
        cdef np.ndarray[np.float_t, ndim=1] alpha_arr
        cdef double margin

build mlpy

Sorry I made you wait later

set LIB=%LIB%;(gsl.lib,cblas.Directory with lib)
python setup.py build_ext --include-dirs=(gsl include directory)
python setup.py install

Or(If you installed anaconda)
python setup.py build_ext --include-dirs=(gsl include directory) --compiler=msvc install

It is completed with. Thank you for your hard work.