Many optimization problems in statistics and machine learning involve continuous parameters. For example, maximum likelihood estimation involves optimizing a log-likelihood function over a continuous domain, possibly with constraints. Recently, however, I had to solve an optimization problem for which the solution vector was a 0/1 binary variable. To solve the
I previously wrote about one way to solve the partition problem in SAS. In the partition problem, you divide (or partition) a set of N items into two groups of size k and N-k such that the sum of the items' weights is the same in each group. For example,
The partition problem has many variations, but recently I encountered it as an interactive puzzle on a computer. (Try a similar game yourself!) The player is presented with an old-fashioned pan-balance scale and a set of objects of different weights. The challenge is to divide (or partition) the objects into
One of the strengths of the SAS/IML language is its flexibility. Recently, a SAS programmer asked how to generalize a program in a previous article. The original program solved one optimization problem. The reader said that she wants to solve this type of problem 300 times, each time using a
An important application of nonlinear optimization is finding parameters of a model that fit data. For some models, the parameters are constrained by the data. A canonical example is the maximum likelihood estimation of a so-called "threshold parameter" for the three-parameter lognormal distribution. For this distribution, the objective function is
When solving optimization problems, it is harder to specify a constrained optimization than an unconstrained one. A constrained optimization requires that you specify multiple constraints. One little typo or a missing minus sign can result in an infeasible problem or a solution that is unrelated to the true problem. This
This article shows how to perform an optimization in SAS when the parameters are restricted by nonlinear constraints. In particular, it solves an optimization problem where the parameters are constrained to lie in the annular region between two circles. The end of the article shows the path of partial solutions
Data analysts often fit a probability distribution to data. When you have access to the data, a common technique is to use maximum likelihood estimation (MLE) to compute the parameters of a distribution that are "most likely" to have produced the observed data. However, how can you fit a distribution
When you run an optimization, it is often not clear how to provide the optimization algorithm with an initial guess for the parameters. A good guess converges quickly to the optimal solution whereas a bad guess might diverge or require many iterations to converge. Many people use a default value
Did you know that you can get SAS to compute symbolic (analytical) derivatives of simple functions, including applying the product rule, quotient rule, and chain rule? SAS can form the symbolic derivatives of single-variable functions and partial derivatives of multivariable functions. Furthermore, the derivatives are output in a form that
