其实很简单，在git bash中，输入：
ALI@ALI-PC MINGW64 /d/SoftWare/Python/Python37/Scripts
$ pip install numpy
$ pip install matplotlib
看到Successfully installed，说明OK。
注：前面的方法，在安装时，遇到了问题，所以请参考下面的方法：
实际在使用Scikit-learn时，出错了，后来发现是numpy+mkl和Scipy没安装。
安装Scikit-learn之前，先装Numpy\Scipy\Matplotlib包，再安装Scikit-Learn。
去到官网下载相应版本：
http://www.lfd.uci.edu/~gohlke/pythonlibs
其中cp27表示CPython 2.7版本，cp34表示CPython 3.4，win_arm64指的是64位版本。
参考：https://blog.csdn.net/hujiuding/article/details/78309309
为了避免冲突， numpy scipy matplotlib 等都应该在官网下载。
值得注意的是：
Scikit-learn integrates classic machine learning algorithms.
Requires numpy+mkl.
Scikit-Learn下载的小节，提示说需要 numpy+mkl.
如果你安装的numpy不含mkl，那么Scikit-Learn可能会报错。
因此，建议是在官网都下载相应文件。
另外，numpy+mkl 比较大，我的浏览器下载速度很慢，我是用百度云盘先离线下载后，再下到本地的。
云工具可参考：https://blog.csdn.net/lemonix7/article/details/
另外值得注意的是，在安装numpy时，
下载到的文件名是：
numpy-1.15.1 mkl-cp37-cp37m-win_amd64.whl
而官方的文件名是（多了一个加号）
numpy‑1.15.1+mkl‑cp37‑cp37m‑win_amd64.whl
自己记得改一下名。
安装成功后，以下import就不会报错了。
from sklearn import linear_model最终的安装日志如下，供参考：
ALI@ALI-PC MINGW64 /d/SoftWare/Python/Python37/Scripts/lib
$ ll
total 250904
-rw-r--r-- 1 ALI 197121 8734219 九月 8 23:55 matplotlib-2.2.3-cp37-cp37m-win_amd64.whl
-rw-r--r-- 1 ALI 197121 230383926 九月 9 01:18 numpy-1.15.1+mkl-cp37-cp37m-win_amd64.whl
-rw-r--r-- 1 ALI 197121 4481636 九月 8 13:55 scikit_learn-0.19.2-cp37-cp37m-win_amd64.whl
-rw-r--r-- 1 ALI 197121 13313205 九月 8 23:54 scipy-1.1.0-cp37-cp37m-win_amd64.whl
-rw-r--r-- 1 ALI 197121 0 九月 8 23:16 注：该文件夹是用于存放从官网下载的whl文件.txt
ALI@ALI-PC MINGW64 /d/SoftWare/Python/Python37/Scripts/lib
$ pip install numpy-1.15.1+mkl-cp37-cp37m-win_amd64.whl
Processing d:\software\python\python37\scripts\lib\numpy-1.15.1+mkl-cp37-cp37m-win_amd64.whl
Installing collected packages: numpy
Successfully installed numpy-1.15.1+mkl
ALI@ALI-PC MINGW64 /d/SoftWare/Python/Python37/Scripts/lib
$ pip install matplotlib-2.2.3-cp37-cp37m-win_amd64.whl
Processing d:\software\python\python37\scripts\lib\matplotlib-2.2.3-cp37-cp37m-win_amd64.whl
Requirement already satisfied: kiwisolver>=1.0.1 in d:\software\python\python37\lib\site-packages (from matplotlib==2.2.3) (1.0.1)
Requirement already satisfied: pyparsing!=2.0.4,!=2.1.2,!=2.1.6,>=2.0.1 in d:\software\python\python37\lib\site-packages (from matplotlib==2.2.3) (2.2.0)
Requirement already satisfied: python-dateutil>=2.1 in d:\software\python\python37\lib\site-packages (from matplotlib==2.2.3) (2.7.3)
Requirement already satisfied: cycler>=0.10 in d:\software\python\python37\lib\site-packages (from matplotlib==2.2.3) (0.10.0)
Requirement already satisfied: numpy>=1.7.1 in d:\software\python\python37\lib\site-packages (from matplotlib==2.2.3) (1.15.1+mkl)
Requirement already satisfied: pytz in d:\software\python\python37\lib\site-packages (from matplotlib==2.2.3) (2018.5)
Requirement already satisfied: six>=1.10 in d:\software\python\python37\lib\site-packages (from matplotlib==2.2.3) (1.11.0)
Requirement already satisfied: setuptools in d:\software\python\python37\lib\site-packages (from kiwisolver>=1.0.1->matplotlib==2.2.3) (40.2.0)
Installing collected packages: matplotlib
Successfully installed matplotlib-2.2.3
ALI@ALI-PC MINGW64 /d/SoftWare/Python/Python37/Scripts/lib
$ pip install scipy-1.1.0-cp37-cp37m-win_amd64.whl
Processing d:\software\python\python37\scripts\lib\scipy-1.1.0-cp37-cp37m-win_amd64.whl
Requirement already satisfied: numpy>=1.8.2 in d:\software\python\python37\lib\site-packages (from scipy==1.1.0) (1.15.1+mkl)
Installing collected packages: scipy
Successfully installed scipy-1.1.0
ALI@ALI-PC MINGW64 /d/SoftWare/Python/Python37/Scripts/lib
$ pip install scikit_learn-0.19.2-cp37-cp37m-win_amd64.whl
Processing d:\software\python\python37\scripts\lib\scikit_learn-0.19.2-cp37-cp37m-win_amd64.whl
Installing collected packages: scikit-learn
Successfully installed scikit-learn-0.19.2
ALI@ALI-PC MINGW64 /d/SoftWare/Python/Python37/Scripts/lib
$