5. Learned Parameter Equation Regression

Equation Learning on the Learned Parameter Networks using LASSO Regression

  In order to obtain an approximate ODE system that models our system, we must infer equations that describe the parameters in the system. At this stage in the process, we've already trained neural networks on the data under our governing dynamical system, but now we how non-interpretable black box functions (the trained parameter networks) that we wish to interpret. This is the equation learning step of the process. We construct a feature space of all the possible components of our function/parameter and use sparse regression techniques to create a linear combination of terms up to a certain degree to estimate the parameter networks.

Sparse Regression

We utilize LASSO regression as our method of sparse linear regression.

• Our Algorithm

Code

• DRUMS_LASSO.py

• Storing and loading results