jupyter出现Intel MKL FATAL ERROR: Cannot load libmkl_intel_thread.dylib的解决办法

版权声明：本文为博主原创文章，转载需要注明来源 https://blog.csdn.net/scheezer/article/details/83791923

背景

因为conda清华的源老旧，pandas还停留在0.20，当前版本已经到了0.23
将清华的源切换回了默认源。

切换后使用conda update --all更新了全部包

问题

切换后运行jupyter，随意执行一行代码即报错：服务似乎挂掉了,但是会立刻重启的
在命令行中提示：Intel MKL FATAL ERROR: Cannot load libmkl_intel_thread.dylib.

解决

查了不少资料，有的说需要使用

conda update conda 

然而使用后无效。

后参考 https://stackoverflow.com/questions/36659453/intel-mkl-fatal-error-cannot-load-libmkl-avx2-so-or-libmkl-def-so

执行

conda install nomkl numpy scipy scikit-learn numexpr 

提示

conda install nomkl numpy scipy scikit-learn numexpr Solving environment: done  ## Package Plan ##    environment location: /Users/wangyu/anaconda3    added / updated specs:     - nomkl     - numexpr     - numpy     - scikit-learn     - scipy   The following packages will be downloaded:      package                    |            build     ---------------------------|-----------------     mkl_fft-1.0.1              |   py36h917ab60_0         125 KB  defaults     scikit-learn-0.20.0        |   py36hebd9d1a_1         5.4 MB  defaults     numpy-base-1.15.3          |   py36ha711998_0         4.0 MB  defaults     libopenblas-0.3.3          |       hdc02c5d_3         8.4 MB  defaults     mkl-service-1.1.2          |   py36h6b9c3cc_4          10 KB  defaults     numexpr-2.6.8              |   py36hafae301_0         125 KB  defaults     nomkl-3.0                  |                0          48 KB  defaults     numpy-1.15.3               |   py36h926163e_0          35 KB  defaults     blas-1.0                   |         openblas          48 KB  defaults     mkl_random-1.0.1           |   py36h78cc56f_0         346 KB  defaults     scipy-1.1.0                |   py36h1a1e112_1        15.4 MB  defaults     ------------------------------------------------------------                                            Total:        33.9 MB  The following NEW packages will be INSTALLED:      libopenblas:  0.3.3-hdc02c5d_3      defaults     nomkl:        3.0-0                 defaults  The following packages will be UPDATED:      blas:         1.0-mkl               defaults --> 1.0-openblas          defaults     numexpr:      2.6.8-py36h1dc9127_0  defaults --> 2.6.8-py36hafae301_0  defaults     numpy:        1.15.3-py36h6a91979_0 defaults --> 1.15.3-py36h926163e_0 defaults     numpy-base:   1.15.3-py36h8a80b8c_0 defaults --> 1.15.3-py36ha711998_0 defaults     scikit-learn: 0.20.0-py36h4f467ca_1 defaults --> 0.20.0-py36hebd9d1a_1 defaults     scipy:        1.1.0-py36h28f7352_1  defaults --> 1.1.0-py36h1a1e112_1  defaults  The following packages will be DOWNGRADED:      mkl-service:  1.1.2-py36h6b9c3cc_5  defaults --> 1.1.2-py36h6b9c3cc_4  defaults     mkl_fft:      1.0.6-py36hb8a8100_0  defaults --> 1.0.1-py36h917ab60_0  defaults     mkl_random:   1.0.1-py36h5d10147_1  defaults --> 1.0.1-py36h78cc56f_0  defaults  

确认安装后再运行jupyter，执行无误。

考虑问题出在mkl-service上面，回滚后即解决问题。

文章来源: jupyter出现Intel MKL FATAL ERROR: Cannot load libmkl_intel_thread.dylib的解决办法