最近在用dtw，发现：
dtw.distance 就不会报错
dtw.distance_matrix_fast 就会报错：
The compiled dtaidistance C library is not available.
See the documentation for alternative installation options.
上dtaidistance的官网看，官网地址：
https://dtaidistance.readthedocs.io/en/latest/usage/installation.html#from-source
叫我尝试其他的安装方法。
官网一共提供了三种这个依赖的安装方法：
- From PyPI
- From Github
- From source
我都尝试了以下，效果都不是很好，还是有那个问题。不过第三种安装方式其实是可以解决问题的，只是我不知道安装目录应该怎么安排。
但是我的另一台电脑不报错，我就对比了以下两个工程Lib/site-packages目录下的dtaidistance，发现，报错的工程这个目录下只有26个文件，不报错的工程下这个目录有34个文件。
所以我怀疑可能是这个包缺少文件导致的。
另外，我点进dtw.distance_matrix_fast 这个函数看到：
def distance_matrix_fast(s, max_dist=None, max_length_diff=None,
window=None, max_step=None, penalty=None, psi=None,
block=None, compact=False, parallel=True, use_mp=False,
only_triu=False):
"""Same as :meth:`distance_matrix` but with different defaults to choose the
fast parallized C version (use_c = True and parallel = True).
This method uses the C-compiled version of the DTW algorithm and uses parallelization.
By default this is the OMP C parallelization. If the OMP functionality is not available
the parallelization is changed to use Python's multiprocessing library.
"""
_check_library(raise_exception=True, include_omp=False)
if not use_mp and parallel:
try:
_check_library(raise_exception=True, include_omp=True)
except Exception:
use_mp = True
return distance_matrix(s, max_dist=max_dist, max_length_diff=max_length_diff,
window=window, max_step=max_step, penalty=penalty, psi=psi,
block=block, compact=compact, parallel=parallel,
use_c=True, use_mp=use_mp, show_progress=False, only_triu=only_triu)
这个函数有个check library，进而觉得是这里有问题，再点进去可以看到：
def _check_library(include_omp=False, raise_exception=True):
if dtw_cc is None:
msg = "The compiled dtaidistance C library is not available.\n" + \
"See the documentation for alternative installation options."
logger.error(msg)
if raise_exception:
raise Exception(msg)
if include_omp and (dtw_cc_omp is None or not dtw_cc_omp.is_openmp_supported()):
msg = "The compiled dtaidistance C-OMP library "
if dtw_cc_omp and not dtw_cc_omp.is_openmp_supported():
msg += "indicates that OpenMP was not avaiable during compilation.\n"
else:
msg += "is not available.\n"
msg += "Use Python's multiprocessing library for parellelization (use_mp=True).\n" + \
"Call dtw.try_import_c() to get more verbose errors.\n" + \
"See the documentation for alternative installation options."
logger.error(msg)
if raise_exception:
raise Exception(msg)
第一个if后面提示的信息和我的报错信息一模一样。至此算是找到了问题：缺少dtw_cc。
于是我把dtaidistance里的dtw_cc.c dtw_cc.pyx都复制到Lib/site-packages目录下，运行发现还是不行。
我又尝试把官网from source的安装方法中的dtaidistance整个文件夹cut出来，贴在Lib/site-packages目录下，替换掉原来的dtaidistance，再运行就成功了。
但是还有以下提示：
The compiled dtaidistance C-OMP library indicates that OpenMP was not avaiable during compilation.
Use Python's multiprocessing library for parellelization (use_mp=True).
Call dtw.try_import_c() to get more verbose errors.
See the documentation for alternative installation options.
不过这个不影响，还是可以计算出结果的。
总结：去看源码来看错误原因。