Windows上安装Tensorflow 以及集成Spark遇到的问题

Windows上安装Tensorflow


参考博客： http://blog.youkuaiyun.com/u013713117/article/details/54616460


第一种安装方式：使用Pip
1： 安装python3.5.x


2： 升级 pip


python -m pip install --upgrade pip




3：安装tensorflow


pip install --upgrade --ignore-installed https://storage.googleapis.com/tensorflow/windows/cpu/tensorflow-0.12.0rc0-cp35-cp35m-win_amd64.whl


> pip install tensorflow
# 或


我用的这个版本：
pip install --upgrade https://storage.googleapis.com/tensorflow/windows/cpu/tensorflow-1.4.0rc0-cp35-cp35m-win_amd64.whl
旧版本：
pip install --upgrade https://storage.googleapis.com/tensorflow/windows/cpu/tensorflow-0.12.0rc0-cp35-cp35m-win_amd64.whl




GPU版本


> pip install tensorflow-gpu  # 需CUDA 8
# 或
> pip install --upgrade https://storage.googleapis.com/tensorflow/windows/gpu/tensorflow_gpu-0.12.0rc0-cp35-cp35m-win_amd64.whl


第二种安装方式： 本地安装方式


下载安装 pandas-0.21.0rc1-cp35-cp35m-win_amd64.whl (md5)


pip install --upgrade   E:\Python\Python-lib\pandas-0.21.0rc1-cp35-cp35m-win_amd64.whl   


pip install --upgrade   E:\Python\Python-lib\pandas-0.21.0rc1-cp27-cp27m-win_amd64.whl 


gensim-3.0.1-cp35-cp35m-win_amd64.whl


先安装：
pip install --upgrade   E:\Python\Python-lib\scipy-1.0.0rc2-cp35-none-win_amd64.whl


后安装：
pip install --upgrade gensim




pip install --upgrade   E:\Python\Python-lib\matplotlib-2.1.0-cp35-cp35m-win_amd64.whl


先安装
pip install --upgrade   E:\Python\Python-lib\Twisted-17.9.0-cp35-cp35m-win_amd64.whl
后安装
pip install --upgrade   E:\Python\Python-lib\Scrapy-1.4.0-py2.py3-none-any.whl


pip install --upgrade   E:\Python\Python-lib\opencv_python-3.3.0+contrib-cp35-cp35m-win_amd64.whl




4： 验证
安装成功后运行


>import tensorflow as tf




如果没有勾选Download debug binaries (requires VS 2015 or later )，报错：


Traceback (most recent call last):
  File "C:\Users\mark\AppData\Local\Programs\Python\Python35\lib\site-packages\tensorflow\python\pywrap_tensorflow.py", line 19, in swig_import_helper
    return importlib.import_module(mname)
  File "C:\Users\mark\AppData\Local\Programs\Python\Python35\lib\importlib\__init__.py", line 126, in import_module
    return _bootstrap._gcd_import(name[level:], package, level)
  File "<frozen importlib._bootstrap>", line 986, in _gcd_import
  File "<frozen importlib._bootstrap>", line 969, in _find_and_load
  File "<frozen importlib._bootstrap>", line 958, in _find_and_load_unlocked
  File "<frozen importlib._bootstrap>", line 666, in _load_unlocked
  File "<frozen importlib._bootstrap>", line 577, in module_from_spec
  File "<frozen importlib._bootstrap_external>", line 906, in create_module
  File "<frozen importlib._bootstrap>", line 222, in _call_with_frames_removed
ImportError: DLL load failed: The specified module could not be found.


解决方法： 
Install Microsoft Visual C++ 2015 Redistributable Update 3 x64.
地址：
https://www.microsoft.com/en-us/download/details.aspx?id=53587






Python集成Spark安装配置：


第一种方法：


windows7 spark单机环境搭建及pycharm访问spark
http://www.cnblogs.com/porco/p/4895421.html




windows7 spark单机环境搭建


follow this link how to run apache spark on windows7
pycharm 访问本机 spark


安装py4j
pip install py4j


配置pycharm
在PYTHON_HOME\lib\site-packages下新建pyspark.pth文件内容为:
D:\program\spark-1.5.1-bin-hadoop2.4\python


试运行


第二种方法：
pip install pyspark




 


ImportError: cannot import name downsample


在学习使用  TensorFlow 学习卷积神经网络的时候，遇到这个问题：


from theano.tensor.signal.pool import downsample


ImportError: cannot import name downsample


解决方案来自大牛的回复 
[http://blog.youkuaiyun.com/u012162613/article/details/43277187#comments ]： 
在较早的版本中，可能会使用from theano.tensor.signal.downsample import max_pool_2d。
根据运行提示，更新到 from theano.tensor.signal.pool import pool_2d 中了。


查看了一下theano下面的包，发现在signal下面没有了downsample了，只有pool和conv了。
如果以后再用downsample的时候，很可能就是调用这两个包里面的东西，改一下就好了。感叹一下，计算机变的真快。


把 from theano.tensor.signal.pool import downsample注释掉,改为


from theano.tensor.signal.pool import pool_2d