How to install Tensorflow on Python 2.7 on Windows?

If you only need TensorFlow because of Keras and your are on Python 2.7.x, you can avoid installing Tensorflow(Google) and replace it by CNTK(Microsoft). According to Jeong-Yoon Lee CNTK is a lot (about 2 to 4 times) faster than TensorFlow for LSTM (Bidirectional LSTM on IMDb Data and Text Generation via LSTM), while speeds for other type of neural networks are close to each other.
Your Keras code does not need to be modified (I checked it with 2 examples of Keras using TensorFlow and succesfully replaced TensorFlow with CNTK, without changing anything the Keras code.

So how do you install it?

-CPU-only version of CNTK:

pip install
https://cntk.ai/PythonWheel/CPU-Only/cntk-2.4-cp27-cp27m-win_amd64.whl

-GPU version of CNTK:

pip install
https://cntk.ai/PythonWheel/GPU/cntk-2.4-cp27-cp27m-win_amd64.whl

-Test CNTK install:

python -c “import cntk; print(cntk.version)”

-Install Keras: The Python Deep Learning library

pip install keras

-Enable CNTK as Keras back end iso TensorFlow

modify the “keras.json” file under %USERPROFILE%/.keras

{
    "epsilon": 1e-07, 
    "image_data_format": "channels_last", 
    "backend": "cntk", 
    "floatx": "float32" 
}

Leave a Comment