Re: [R] genoud problem

Hi Shubha, genoud does not return the initial fit value. But you could easily obtain it by passing your starting values to your function directly. Alternatively, one can have genoud print out the entire initial population (or the entire population as is evolves), and one can then decide to

Re: [R] Increasing precision of rgenoud solutions

Hi Paul, Solution.tolerance is the right way to increase precision. In your example, extra precision *is* being obtained, but it is just not displayed because the number of digits which get printed is controlled by the options(digits) variable. But the requested solution precision is in the

Re: [R] Increasing precision of rgenoud solutions

Hi Paul, I see. You want to increase the population size (pop.size) option---of lesser importance are the max.generations, wait.generations and P9 options. For more details, see http://sekhon.berkeley.edu/papers/rgenoudJSS.pdf. For example, if I run a - genoud(myfunc, nvars=2,

Re: [R] Bad optimization solution

The issue is that you are using a derivative based optimizer for a problem for which it is well known that such optimizers will not perform well. You should consider using a global optimizer. For example, rgenoud combines a genetic search algorithm with a BFGS optimizer and it works well for

Re: [R] The confidence level of p-value of ks.boot

Hi Gala, The default p-value is the bootstrap p-value for the ks-test. Bootstrapping is highly recommended because the bootstrapped Kolmogorov-Smirnov test, unlike the standard test, provides correct coverage even when there are point masses in the distributions being compared. The bootstrap

Re: [R] multinomial logistic regression with equality constraints?

As we noted earlier and as is clearly stated in the docs, multinomRob is estimating an OVERDISPERSED multinomial model. And in your models here the overdispersion parameter is not identified; you need more observations. Walter pointed out using the print.level trick to get the coefs for the

Re: [R] multinomial logistic regression with equality constraints?

Hi Roger, Yes, multinomRob can handle equality constraints of this type---see the 'equality' option. But the function assumes that the outcomes are multinomial counts and it estimates overdispersed multinomial logistic models via MLE, a robust redescending-M estimator, and LQD which is another

Re: [R] multinomial logistic regression with equality constraints?

Mebane Jasjeet Singh Sekhon writes: Hi Roger, Yes, multinomRob can handle equality constraints of this type---see the 'equality' option. But the function assumes that the outcomes are multinomial counts and it estimates overdispersed

Re: [R] ks.test not working?

cannot compute correct p-values with ties in: ks.test(x, pgev, fit$mle[1], fit$mle[2], fit$mle[3]) You may want to use the ks.boot function in the Matching package which implements a bootstrap ks-test which provides consistent pvalues (achieved significance levels) when there are ties.

Re: [R] Var.calc in Match()

Does anyone else find that using the Var.calc option (for heteroscedasticity consistent std. errors) in Match() (from the Matching library) slows down computation of the matching estimator by a lot? The Var.calc option to Match() slows down the function because an additional loop through the

Re: [R] number of matches when using Match()

How do you go about deciding how many matches you will use? With my data, my standard errors generally get smaller if I use more matches. Generally, select the max number of matches that result in good or acceptable balance (hence bounding bias due to the observed confounders). See the

Re: [R] R performance: different CPUs

Hi, 64bit CPUs, such as opterons, help significantly with large databases or if you are running multiple processes. But there is a speed penalty if you are not. Some packages can make use of multiple processors, such as my rgenoud (genetic optimization using derivatives) and Matching packages,

Re: [R] R performance: different CPUs

=== Thomas Lumley writes: On Wed, 5 Apr 2006, Jasjeet Singh Sekhon wrote: Hi, 64bit CPUs, such as opterons, help significantly with large databases or if you are running multiple processes. But there is a speed penalty if you are not. This would

Re: [R] any more direct-search optimization method in R

Given that information, I think a genetic algorithm should probably do well with your problem. You may want to try the rgenoud package (R-GENetic Optimization Using Derivatives) which is on CRAN. For more information see: http://sekhon.berkeley.edu/rgenoud/ It works well for these kinds of

Re: [R] Parallel computing in R for dummies--how to optimize an external model?

Hi Scott, It is difficult to debug your issue without more information. Would it be possible to email me code of a simple example? Cheers, Jas. === Jasjeet S. Sekhon Associate Professor

[R] [R-pkgs] New Package for Multivariate and Propensity Score Matching

Matching version 0.48 is now available on CRAN. Matching provides functions for estimating causal effects by multivariate and propensity score matching. The package includes a variety of univariate and multivariate tests to determine if balance has been obtained by the matching procedure. These

[R] [R-pkgs] New Package: multinomRob

Author: Walter R. Mebane, Jr. [EMAIL PROTECTED], Jasjeet Singh Sekhon [EMAIL PROTECTED] Maintainer: Jasjeet Singh Sekhon [EMAIL PROTECTED] Description: overdispersed multinomial regression using robust (LQD and tanh) estimation Depends: R (= 1.7.0), rgenoud (= 1.22), MASS (= 7.1-8), mvtnorm (= 0.6-3