johannbrehmer/madminer: v0.2.8

Breaking / API changes:

• The keyword trainer in MLForge.train() has been renamed to optimizer
• The gradient of network output wrt x is no longer calculated, the corresponding keywords in MLForge.evaluate() and MLForge.train() are gone
• MLForge has new methods evaluate_log_likelihood(), evaluate_log_likelihood_ratio(), and evaluate_score(); MLForge.evaluate() just wraps around them now

• Fixed a number of bugs affecting neural density estimation methods ("NDE" and "SCANDAL")
• Fixed bugs when trying to use CASCAL

• Rewrote training procedure from scratch
• Some more refactoring of the ML parts