How to deal with TensorFlow contrib modules?

Question

Google

The TensorFlow 1.3 release has all sorts of high-level modules: Layers, Estimators, Classifiers. They all have dry API documentation , where methods and parameters are simply described.
<баттхёрт>Here, a new person "from the street" has come, and how can he, *#%&, sort out this wealth? There are dofigischi of these classes, and there is no overall picture. At most, there are “ guides ” teasingly named , which are not guides at all, but simply links to the same API dock pages grouped by topic. </баттхёрт>
Books, videos, lessons - all appear with a delay. And not all topics are covered in detail.
Here, there is an "official" example of the implementation of the MNIST classifier using random forest. Not only does the code no longer work with TF1.3 / Python3 - it swears at an unusable typeuint8somewhere out there, in the wilds of TF. But I can't find descriptions of the magical passes used there! The tensor forest
API doc is just a spit in the soul - one line "Random forest implementation in tensorflow." Everything! For example, why do they pass some kind of parameter instead of a dictionary , about which there are no mentions in the docs at all. Friends who have already eaten the dog, shine a flashlight, where to look, what to do, what to do?
ForestHParams

Answer 1

so, as with the usual pythonic
dir () and forward
, it is clear that you need to know the subject area, but you can dig up to the source