Hi R-helpers, Can some one tell me how to train 'mynn' of this type?: mynn <- nnet(y ~ x1 + ..+ x8, data = lgist, size = 2, rang = 0.1, decay = 5e-4, maxit = 200) I assume that this nn is untrained, and to train I have to split the original data into train:test data set, do leave-one-out refitting to refine the weights (please straighten this up if I was wrong). I just don't know how to do it in R. Is 'training' and 'training.reports' in (AMORE) able to do it? Thank you for any light on this. Ilh
On Dec 14, 2007 10:52 AM, G Ilhamto <gilhamto at gmail.com> wrote:> Hi R-helpers, > > Can some one tell me how to train 'mynn' of this type?: > mynn <- nnet(y ~ x1 + ..+ x8, data = lgist, size = 2, rang = 0.1, > decay = 5e-4, maxit = 200) >nnet will estimate model parameters for the model that you have specified. If you want to understand what parameters (e.g. size, decay) should be used, the train function in the caret package uses a variety of resampling methods to help pick those parameters (and will refit the model based on that). There are similar functions in other packages (like e1071). Install the caret package and use vignette("caretTrain") to see the details. -- Max
Dear Ilh I was planning to dedicate a few hours this christmas to improve the AMORE package. I would like to explore the possibilities of the openMP library and to programm a few more training algorithms. Right now the rbf extension is ready but under testing. I think that what you ask for would not need much time to get done. If you provide me a detailed definition of your needs I can include them in my to-do list. For that purpose it could be adequate to send me an email to mcasl at unileon.es Best wishes. Manuel> Hi R-helpers, > > Can some one tell me how to train 'mynn' of this type?: > mynn <- nnet(y ~ x1 + ..+ x8, data = lgist, size = 2, rang = 0.1, > decay = 5e-4, maxit = 200) > > I assume that this nn is untrained, and to train I have to split the > original data into train:test data set, > do leave-one-out refitting to refine the weights (please straighten > this up if I was wrong). > > I just don't know how to do it in R. Is 'training' and > 'training.reports' in (AMORE) able to do it? > > Thank you for any light on this. > > Ilh > > >-- Manuel Castej?n Limas ?rea de Proyectos de Ingenier?a Departamento de Ingenier?as Mec?nica, Inform?tica y Aeroespacial Universidad de Le?n. e-mail: manuel.castejon en unileon.es Escuela de Ingenier?as Industrial e Inform?tica Campus de Vegazana, sn. 24007. Le?n