# Choose your adventure

# Choose your adventureΒΆ

At this stage we will be training a neural network to approximate a *likelihood* or a *likelihood ratio*
for down-stream statistical analysis. The papers mentioned in the Introduction
describe these techniques in detail. There are basically three options for this step:

An approximate likelihood ratio (

**recommended**for the tutorial, ~20 min).An approximate score, which is like the optimal observables (

**stretch goal**for the tutorial).An approximate likelihood (like a high-dimensional histogram, but with a neural network).

This is where the machine learning training happens, and MadMiner uses Pytorch.