# init repo notebook
!git clone https://github.com/rramosp/ppdl.git > /dev/null 2> /dev/null
!mv -n ppdl/content/init.py ppdl/content/local . 2> /dev/null
!pip install -r ppdl/content/requirements.txt > /dev/null

5. Similarity of distributions

• KL divergence

• Wassesrtein divergen

• Why log (information theory + convenient algebra + numerically better + ok in most arg max/min situations since it is monotonous + when assuming gaussians cancels out exponential)

• Questions:

  • How similar are two images with similar pixel distributions? –> maybe the things we what to deal with probabiities are not pixels, but something else (objects, full imgs, etc.)