[MATH-1009] PolynomialFitter

• Type: Improvement
• Status: Closed
• Priority: Minor
• Resolution: Fixed
• Affects Version/s: 3.2
• Fix Version/s: 3.3

Environment

All

• Assignee: Unassigned
• Reporter: Konstantin Berlin
• Created: 18/Jul/13 13:04
• Updated: 19/May/14 15:13
• Resolved: 21/Feb/14 11:32

Description

org.apache.commons.math3.fitting.PolynomialFitter

should be implemented using linear least-squares method like QR decomposition solver.

There are several reasons for this
1) Nonlinear methods are much slower
2) Linear methods (QR and SVD) are numerically more stable.
3) By storing the QR decomposition it is possible to recompute the solution for different input data values.

See
http://mathworld.wolfram.com/LeastSquaresFittingPolynomial.html

Issue Links

• is related to: MATH-1099 Improve speed + accuracy in GaussNewtonOptimizer
• relates to: MATH-1014 Remove optimizer from constructor of "CurveFitter" subclasses