The main objective of this release is improving and adding inference routines, such as conditional sampling and MAP, and add conditional circuit. These are very useful to implement semantic probabilistic layers. Also, we aim at implementing circuit templates as to represent a set of (hierarchical) tensor factorizations.

Features:

• implement MAP inference in monotonic circuits
• refactor sampling, and support conditional sampling
• implement conditional circuits
  • maybe dynamic compilation rules based on the parameterizations
• make sure copy.deepcopy works
• implement circuit pipelines serialization

[done] implement tensor factorization templates
[done] replace the concept of channels with variables