Due to the necessity of including paramter estimation within the SINDy approach the sparse optimization should be reiterated again. Since this requires an alternating update of the evaluated basis function with all of the problem data, the optimizers should either get a Sparse Regression problem as a base problem or a cache storing the basis, jacobian, data driven problem, and possible cached values for efficient computation. Also options like denoise and normalize need to move. This would in term enable a more efficient solving via a step! function. I will add a protoype for STLSQ soonish in a Draft.
Due to the necessity of including paramter estimation within the SINDy approach the sparse optimization should be reiterated again. Since this requires an alternating update of the evaluated basis function with all of the problem data, the optimizers should either get a Sparse Regression problem as a base problem or a cache storing the basis, jacobian, data driven problem, and possible cached values for efficient computation. Also options like denoise and normalize need to move. This would in term enable a more efficient solving via a step! function. I will add a protoype for STLSQ soonish in a Draft.