FormaK aims to combine symbolic modeling for fast, efficient system modelling
with code generation to create performant code that is easy to use.
The Five Key Elements the library provides to achieve this user experience are:
Python Interface to define models
Python implementation of the model and supporting tooling
Integration to scikit-learn to leverage the model selection and parameter tuning functions
C++ and Python to C++ interoperability for performance
C++ interfaces to support a variety of model uses
This design provides an extension to the second of the Five Keys "Python
implementation of the model..." and the fifth of the Five Keys "C++ interfaces
to support a variety of model uses" to support the performance goals of the
library: Common Subexpresion Elimination (CSE).
CSE is a transformation on the compute graph of the underlying model (for
either Python or C++) to remove duplicate compuation at multiple scales. For
example, if a sensor model computes a + b multiple times, CSE identifies this
subexpression and computes it once. This could also apply to a more complicated
expression like 9.81 * sin(theta + bias).
One of the nicest benefits of this transformation is that it will provide a
performance benefit to the model without a compromise to the user interface.
FormaK aims to combine symbolic modeling for fast, efficient system modelling with code generation to create performant code that is easy to use.
The Five Key Elements the library provides to achieve this user experience are:
This design provides an extension to the second of the Five Keys "Python implementation of the model..." and the fifth of the Five Keys "C++ interfaces to support a variety of model uses" to support the performance goals of the library: Common Subexpresion Elimination (CSE).
CSE is a transformation on the compute graph of the underlying model (for either Python or C++) to remove duplicate compuation at multiple scales. For example, if a sensor model computes
a + bmultiple times, CSE identifies this subexpression and computes it once. This could also apply to a more complicated expression like9.81 * sin(theta + bias).One of the nicest benefits of this transformation is that it will provide a performance benefit to the model without a compromise to the user interface.