robertostling / efmaral Goto Github PK

"python3 align.py -i data.txt -v > test.moses --output-prob prob.txt"
I run this, but only generates the file"prob1.txt" , there is nothing in test.moses( it should be the alignment result but the content in the file is empty). And by the way, I also tried eflomal , it seems that it can only generate sentence_score, I don't know whether eflomal can also generate alignment probabilities like efmaral. Thank you for your kindness and patience.

Is efmaral supported on Windows? I can't find any relevant instructions.

Thanks for the very efficient and accurate aligner. I would be great to have a flag that allows a user to indicate whether he or she wants the input to be lowercased or not :) It seems like the input is always lowercased (correct me if I am wrong).

I added the "--output-prob" parameter, but I can't figure out how to use it. How can I get word alignment results with probability? (like "word1 word2 0.01")

This is a problem in setup.py and is caused by Cython not having the correct path to numpy includes if numpy is in a virtual env. A solution is to add include_dirs to every Extension which uses numpy. Working setup.py below:

from distutils.core import setup, Extension
from Cython.Build import cythonize
import numpy

gibbsmodule = Extension(
    'gibbs',
    sources=['gibbs.c'],
    libraries=[],
    include_dirs=[numpy.get_include()],
    # NOTE: the -Wno.. arguments are to compensate for a bug in the build
    # system
    extra_compile_args=['-std=c99', '-Wall', '-fopenmp',
                        '-Wno-error=declaration-after-statement',
                        '-Wno-declaration-after-statement',
                        '-Wno-unused-function',
                        # Enable this to use simd_math_primes.h for expf/logf
                        # This should speed up fertility distribution sampling
                        # by about 20%, so overall impact is quite limited.
                        #'-DAPPROXIMATE_MATH',
                        ],
    extra_link_args=['-lgomp'])

cyalign_ext=Extension('cyalign',['cyalign.pyx'],include_dirs=[numpy.get_include()])

setup(
    name = 'Gibbs aligner',
    ext_modules = cythonize(cyalign_ext) + [gibbsmodule]
)

Can you please create a release so I can create a better workflow for EasyBuild.