Software and libraries:
- Anaconda Python 3.7
- Tensorflow
- Keras
- OpenAI Gym
Platform I have tested on: openSUSE Leap 15 (but setup the environment in distributions like Ubuntu or based on Ubuntu might be less painfull to get started quickly)
To download anaconda:
- newest: https://www.anaconda.com/download/#linux
- the one I used: https://repo.continuum.io/archive/Anaconda3-5.3.1-Linux-x86_64.sh
to install:
$ chmod +x Anaconda3-5.3.1-Linux-x86_64.sh
$ ./Anaconda3-5.3.1-Linux-x86_64.sh
$ conda --version
# example output -> conda 4.5.11
Instead of using the default Anaconda environment, we create a separate environment. Let's check the installed media first:
$ conda env list
Example output:
# conda environments:
#
base * /home/faruk/anaconda3
Let's create new anaconda environment:
$ conda create --name deeplearning anaconda python=3.6
Tensorflow supports currently python 3.6 so it is important to create py3.6 environment. "anaconda" is the metapackage that includes all of the Python packages comprising the Anaconda distribution. python=3.6 is the package and version you want to install in this new environment. This could be any package, such as numpy=1.7, or multiple packages.
Once the necessary downloads have been made, let's check if the environment has been installed:
$ conda env list
Example output:
# conda environments:
#
base * /home/faruk/anaconda3
deeplearning /home/faruk/anaconda3/envs/deeplearning
The asterisk sign indicates the currently active environment. To switch to "deeplearning" (to enable this environment), enter the following command:
$ source activate deeplearning
After doing this, you will see (deeplearning) at the beginning of the shell entry. If you enter conda env list again, this time the star will indicate that deeplearning environment is active.
To exit from the environtment:
$ source deactivate
First, if the environment is passive, activate the deeplearning environment. Then enter the following command to install Tensorflow and Keras packages:
$ conda install -c conda-forge keras tensorflow
Tensorflow and Keras packages will be installed. To check if packages are installed, give the following command:
$ conda list | grep 'tensorflow\|keras'
Example output:
keras 2.2.4 py36_0 conda-forge
keras-applications 1.0.4 py_1 conda-forge
keras-preprocessing 1.0.2 py_1 conda-forge
tensorflow 1.10.0 py36_0 conda-forge
You need to install/set up dependencies first.
openSUSE Leap 15:
$ sudo zypper install cmake libav-tools librhash0 libuv1 python3-devel python3-opengl python3-opengl-accelerate python2-opengl python2-opengl-accelerate python3-numpy swig zlib-devel libjpeg8 libjpeg8-devel libjpeg8-32bit libjpeg8-devel-32bit xvfb-run xorg-x11-devel libboost*devel ghc-sdl2 ghc-sdl2-devel python3-glfw python2-glfw libglfw3 libglfw3 glfw2-devel libglfw-devel python3-imageio python2-imageio python-cffi python2-lockfile python3-lockfile python2-Cython python3-Cython python2-Pillow python3-Pillow
To install dependencies on Ubuntu and its variants, use the following command:
$ sudo apt-get install -y python-numpy python-dev libboost-all-dev libsdl2-dev swig cmake zlib1g-dev libjpeg-dev xvfb libav-tools xorg-dev python-opengl
Then, enter the following commands in order to install the Gym package (make sure the deeplearning environment is activated):
$ pip install ImageHash ipdb pytest pytest-instafail scipy sphinx sphinx_rtd_theme numpydoc PyHamcrest
$ pip install 'gym[all]'
current packets can be mismatched between old dependencies. if you cannot solve errors, try to download old packages, try this:
$ pip install gym==0.9.4
To verify that the package is installed, enter the following command:
$ pip freeze | grep gym
At least one of the lines have to be gym. Sample output:
gym==0.9.4
If you get error while trying to run "jupyter notebook" (OSError: [Errno 99] Cannot assign requested address) you can try:
$ jupyter notebook --ip=127.0.0.1