Document installation in Anaconda environment

[mfeurer]

If a user is installing auto-sklearn in an environment where the compiler version and the version of the compiler which was used to compile the python runtime don't work with each other, importing the SMAC random forest will fail. This is for example present in #155, #176, #168, #161, #159, #141, #119, #101. I have pushed an easy workaround in #179 which uses a gcc shipped with anaconda. This easy workaround needs to be document documented so new users have less trouble with the installation.

[Motorrat]

On ubuntu
sudo apt-get install build-essential gcc
resolved the problem

[Motorrat]

pip uninstall xgboost; pip install --no-cache-dir -v xgboost==0.4a30
conda install libgcc 

were needed for xgboost in 0.1.3 #247

[mfeurer]

Documented in #273. Will be merged into development/master.