Reading of the userguide

[FrederikHeber]

@ZofiaTr wrote in #51:

I was also thinking, that it might be very useful if Ben @TeeBee2 could read some chosen parts of the userguide. I suggested some modifications as mentioned in #51 #50 #46 and updated them in the branch Userguide_Samplers, but I might be naturally wrong in some sense, so it would be great to have his opinion. What do you think @FrederikHeber ?

[FrederikHeber]

Again, I absolutely agree that @TeeBee2 should give the guide a reading, too.

However, let me say a few things in defense caused by my premonition on why those tickets are referenced above: The userguide needs to be comprehensible and have an appeal for everyone not just mathematicians with statistical mechanics background and statisticians. I am thinking especially of computer science folks but also people from all fields that use neural networks and machine learning. I have refrained from beating anyone in the guide with the stat.mech. stick unless it serves a purpose. Moreover, I would like this package to be approachable especially by students, be it master or phd level. To this end, I have kept it as simple as possible.

At its current stage of the guide I'll try to give an easy introduction to the questions that are the heart of TATi, namely the shape of the loss manifold. I explain how to use the new simulation interface to take samples from it. I go through implementing a GLA2 sampler first in python (using the simulation module) and this bridges with the GLA2 implementation in the programmer's guide.

I think the better place for in-depth discussions about distributions, averages and unbiased sampling are the upcoming papers.