Fig. 10

Overview of the Infrared architecture. The C++ core engine is connected to a high-end modeling layer in Python by a hybrid connecting layer. The C++ core implements the computational engine to evaluate forward and traceback algorithms on cluster trees as generic code (e.g. supporting different algebras as evaluation policies; PF for partition function, Arctic for maximization). This optimized low-level layer is exposed to Python using pybind11; the core functionality is moreover extended (by tree decomposition, using module treedecomp, population of the cluster tree...) on the Python side to implement the full computational engine. Finally, the high-level modeling interface of Infrared offers functionality to model and solve feature networks, perform sampling targeting specific features (multidimensional Boltzmann sampling), define custom constrain and function types. Moreover, we include the module rna for RNA-specific functionality