Convex piecewise-linear fitting
Summary (1 min read)
1 Convex piecewise-linear fitting problem
- The convex piecewise-linear fitting problem (1) is to find the function f , from the given family F of convex piecewise-linear functions, that gives the best RMS fit to the given data.
- Of course the authors can expand any function with the more general form (4) into its max-affine representation.
- This allows us to normalize the dependent variable data in various ways.
- 4 Outline In Sect. 2 the authors describe several applications of convex piecewise-linear fitting.
- In Sect. 3, the authors describe a basic heuristic algorithm for solving the maxaffine fitting problem (1).
2 Applications
- In this section the authors briefly describe some applications of convex piecewise-linear fitting.
- This convex piecewise-linear approximate value function can be used to construct a simple feedback controller that approximately minimizes fuel use; see, e.g., Bemporad et al. (2002).
- The authors can write the algorithm as LEAST-SQUARES PARTITION ALGORITHM.
- The authors can interpret the algorithm as a Gauss-Newton method for the problem (3).
- In any case, convergence failure has no practical consequences since the algorithm is terminated after some fixed maximum number of steps, and moreover, the authors recommend that it be run from a number of starting points, with the best fit obtained used as the final fit.
4 Numerical examples
- Figure 1 shows the RMS fits obtained after Ntrials = 10 trials (top curve), and after Ntrials = 100 trials (bottom curve).
- Evidently the best of even a modest number of trials will be quite good.
- The authors set the iteration limit for both forms as lmax = 100, and take the best fit obtained in Ntrials = 10 trials.
- Figure 4 shows the RMS fit obtained for the two forms, versus k.
5 Conclusions
- The authors have described a new method for fitting a convex piecewise linear function to a given (possibly large) set of data (with a modest number of independent variables).
- Numerical examples suggest, however, that the method works very well in practice, on data that can be fit well by a convex function.
- Data samples can be used to generate piecewise-linear convex functions, which in turn can be used to construct linear programming models.
- This work was carried out with support from C2S2, the MARCO Focus Center for Circuit and System Solutions, under MARCO contract 2003-CT-888.
- The authors are grateful to Jim Koford for suggesting the problem.
Did you find this useful? Give us your feedback
Citations
1,369 citations
Cites methods from "Convex piecewise-linear fitting"
..., using the methods described in [50]–[52], replacing it with a piecewise-affine function with a manageable number of regions....
[...]
1,215 citations
Cites methods from "Convex piecewise-linear fitting"
...But there is a relatively simple method, based on monomial fitting and data point clustering, that works well in practice [103]....
[...]
271 citations
Cites methods from "Convex piecewise-linear fitting"
...The advantage of the PWL technique of [27] is that it optimally determines the intervals over which the linear segments are defined....
[...]
...Beside the method introduced in [27], there are commercial solvers which are able to solve this type of problems, e....
[...]
...A heuristic least-squares method is proposed in [27] to solve this problem....
[...]
183 citations
130 citations
Cites background or methods from "Convex piecewise-linear fitting"
...Refitting a series of hyperplanes can be done in a frequentist (Magnani and Boyd, 2009) or Bayesian (Hannah and Dunson, 2011) manner....
[...]
...Similar methods for refitting hyperplanes have been proposed in Breiman (1993) and Magnani & Boyd (2009)....
[...]
...8 Magnani and Boyd (2009) 10156....
[...]
...Refitting hyperplanes in this manner can be viewed as a Gauss-Newton method for the non-linear least squares problem (Magnani & Boyd 2009), minimize n∑ i=1 ( yi − max k∈{1,...,K} ( αk + β T k xi ))2 ....
[...]
...Low noise or noise free problems often occur when a highly complicated convex function needs to be approximated by a simpler one (Magnani and Boyd, 2009)....
[...]
References
[...]
33,341 citations
7,015 citations
"Convex piecewise-linear fitting" refers methods in this paper
...The algorithm is closely related to the k-means algorithm used in least-squares clustering (Gersho and Gray 1991)....
[...]
1,215 citations
"Convex piecewise-linear fitting" refers background in this paper
...The function ψ is convex and piecewise-linear (see, e.g., Boyd and Vandenberghe 2004); the function φ is evidently bi-affine in x and (b,B)....
[...]
...Then the problem (1) can be solved, exactly, via a quadratic program (QP); see (Boyd and Vandenberghe 2004, Sect....
[...]
765 citations
"Convex piecewise-linear fitting" refers methods in this paper
...This convex piecewise-linear approximate value function can be used to construct a simple feedback controller that approximately minimizes fuel use; see, e.g., Bemporad et al. (2002)....
[...]
448 citations