Rolling back to previous Dockerfile, cannot get xvfb to work with binder

Dockerfile

@@ -2,59 +2,34 @@ FROM andrewosh/binder-base
 
 USER root
 
-# Extra packages needs for OpenAI gym
 RUN apt-get update -y &&\
     apt-get install --fix-missing -y \
-        cmake\
-        libav-tools\
-        libboost-all-dev\
-        libjpeg-dev\
-        libsdl2-dev\
-        python-dev\
         python-numpy\
-        python-opengl\
-        python-pip\
-        python3-dev\
-        python3-numpy\
-        python3-opengl\
-        python3-pip\
-        swig\
-        xorg-dev\
+        python-dev\
+        cmake\
+        zlib1g-dev\
+        libjpeg-dev\
         xvfb\
-        zlib1g-dev &&\
+        libav-tools\
+        xorg-dev\
+        python-opengl\
+        libboost-all-dev\
+        libsdl2-dev\
+        swig &&\
     apt-get clean &&\
     rm -rf /var/lib/apt/lists/* /tmp/* /var/tmp/*tmp
 
 USER main
 
-# Add nbextensions to Jupyter and activate the Table of Contents
+# Python 2
+RUN conda install -c jjhelmus tensorflow=0.10.0
 RUN conda install -c conda-forge jupyter_contrib_nbextensions
 RUN jupyter contrib nbextension install --user
 RUN jupyter nbextension enable toc2/main
+RUN pip install --upgrade gym
+RUN pip install --upgrade 'gym[atari]'
 
-# Use upgraded pip
-RUN /usr/bin/pip install --upgrade --user pip wheel
-RUN /usr/bin/pip3 install --upgrade --user pip wheel
-ENV PATH /home/main/.local/bin:$PATH
-
-# Install scientific packages
-RUN pip install --upgrade --user matplotlib numexpr numpy pandas Pillow protobuf psutil scipy scikit-learn sympy
-RUN pip3 install --upgrade --user matplotlib numexpr numpy pandas Pillow protobuf psutil scipy scikit-learn sympy
-
-# Install TensorFlow
-RUN pip install --upgrade --user https://storage.googleapis.com/tensorflow/linux/cpu/tensorflow-0.11.0rc0-cp27-none-linux_x86_64.whl
-RUN pip3 install --upgrade --user https://storage.googleapis.com/tensorflow/linux/cpu/tensorflow-0.11.0rc0-cp34-cp34m-linux_x86_64.whl
-
-# Install OpenAI gym
-RUN pip install --upgrade --user gym
-RUN pip3 install --upgrade --user gym
-RUN pip install --upgrade --user "gym[atari]"
-RUN pip3 install --upgrade --user "gym[atari]"
-
-# Replace the default script to start Jupyter with xvfb-run so that OpenAI gym
-# can render some of the environments (eg. CartPole-v0) without crashing when
-# it tries to access the display on a headless server.
-ADD start-notebook.sh /home/main/
-
-RUN ipython2 kernel install --user
-RUN ipython3 kernel install --user
+# Python 3
+RUN conda install -n python3 -c jjhelmus tensorflow=0.10.0
+RUN /bin/bash -c "source activate python3 && pip install --upgrade gym"
+RUN /bin/bash -c "source activate python3 && pip install --upgrade 'gym[atari]'"