I am using Anaconda for Python. Want to use "KERAS" deep learning module into SPYDER. When I am installing "THEANO" AND "KERAS" using conda, I can successfully import THEANO but when I am trying to import "KERAS", it needs "TENSERFLOW" as a dependency. I am struggling to install "TENSORFLOW", without creating a virtual machine (VM). I want to use a single root environment, where "KERAS" will be a head module. I have tried both PIP and CONDA. Looking for the suggestion. I have already searched videos and blog, most of them using VM.
Looking for a good genuine solution.
Many thanks in advance.
(my experience relates to a linux environment, but it should also work for a windows environment)
You can start by creating a new environment for python (such that you can work isolated from your system's python library, and can update to the latest versions of the python libraries).
1) install tools
sudo apt-get install python-pip python-dev python-virtualenv
2) virtualenv ~/deeplearning (or another directory name)
3) source ~/deeplearning/bin/activate.sh
4) "pip install --upgrade tensorflow" or "pip install --upgrade tensorflow-gpu¨
(depending on whether you want to use GPU)
5) sudo pip install keras
Keras uses Tensorflow by default. If you want to use Theano, which I recommend, because it's more stable, you have to modify $HOME/.keras/keras.json.
See
https://keras.io/backend/
You need to modify the backend in keras.json located in $HOME/.keras. Change it to theano
Many thanks for your replies.
I got it installed. The procedure is given as follows:
Installing "TensorFlow" and "Keras" on Windows with Anaconda.
I am going to show you how to install "Keras" a deep learning library available in Python to be used via Anaconda.
1) Install Anaconda.
https://www.continuum.io/downloads#windows (Python 3.6 version with 64 Bit Installer)
2) Open command prompt and type
condaupateconda
3) Create a virtual environment into anaconda for installing keras and tensorflow as backend. It is good to create a virtual environment for each project so you can work separately and version will not conflict.
conda create -n ENVtensorflow python=3.5 anaconda
conda create will create a new environment with name ENVtensorflow with the version of Python 3.5 into the anaconda platform.
4) Now activate your virtual environment
activate ENVtensorflow
5) Now copy and paste the command and URL given below using pip installer
pip install --ignore-installed --upgrade https://storage.googleapis.com/tensorflow/windows/cpu/tensorflow-1.2.0-cp35-cp35m-win_amd64.whl
6) Check you tensorflow installation
Type python on the command prompt and press enter
Type import tensorflow and if no errors appear that means you have successfully installed tensorflow
Type exit() to come out
7) Install keras
In command prompt type pip install keras, then
Type python on the command prompt and press enter
Type import keras and if no errors appear that means you have successfully installed keras
Type exit() to come out
Now you can use "keras".
If anyone is interested in the GPU installation of tensorflow and keras :
https://github.com/antoniosehk/keras-tensorflow-windows-installation
https://www.pyimagesearch.com/2016/11/14/installing-keras-with-tensorflow-backend/
- Badges
- Science topic