使用 Nvidia GPU 在 Windows 10 中设置 TensorFlow 环境

Question

I want to implement this example. And thus I need to install python along with some libraries including Scikit-Learn, Numpy, Scipy, matplotlib.pyplot, Pandas, Keras, TensorFlow on my Windows 10 machine.

Currently, I can not use my GPU with TensorFlow. I tried installing CUDA. But still having difficulties setting path variables for python. I also tried installing Tensorflow with Anaconda. But that didn't help.

May I get a suggestion on installing python and its machine-learning packages on Windows with Nvidia GPU support in a fashion that doesn't have dependency issues?

Answer 1

Install python 3.6. Then use pip to install those packages. pip should be bundled with your Python install.

Anaconda has caused me many issues on windows personally. Try to avoid it if possible in my opinion.

Answer 2

I also install tensorflow recently, I find it's very smoothly. As you have installed vs2015, you can then install cuda. When you install the latest cuda, it will config the environment path and config for vs2015 automatically. After this, install python3.5, then use pip install tensorflow. Then you can run the tesorflow demo, when you encounter path issue, you just add it to the path, I remeber there are very file path issue. And when you use python, sometime you'll encounter module not find. Then just install these modules with pip.