[BioC] Fwd: question regarding to MLInterface

[Vincent Carey]

I had responded privately to this inquiry as I had not seen it at Bioc
mailing list.  Here is my response.


---------- Forwarded message ----------
From: Vincent Carey <stvjc at channing.harvard.edu>
Date: 2010/11/16
Subject: Re: [BioC] question regarding to MLInterface
To: jrwang at itri.org.tw


Let's take a concrete example

example(xvalSpec)

this generates an object nn1cv

> nn1cv
MLInterfaces classification output container
The call was:
MLearn(formula = sp ~ CW + RW, data = crabs, .method = nnetI,
   trainInd = xvalSpec("LOG", 5, balKfold.xvspec(5)), size = 3,
   decay = 0.01)
Predicted outcome distribution for test set:

 B   O
102  98
history of feature selection in cross-validation available; use fsHistory()

it includes results of each CV step, but it is wrapped up fairly
tightly.  You have
to use RObject at two levels, first for the CV-generated object, second at the
CV-step level ... here are the first two steps for the example

> lapply(RObject(nn1cv),function(x)summary(RObject(x$mlans)))
[[1]]
a 2-3-1 network with 13 weights
options were - entropy fitting  decay=0.01
 b->h1 i1->h1 i2->h1
 -8.81  -1.33   5.08
 b->h2 i1->h2 i2->h2
 0.01   0.30   0.12
 b->h3 i1->h3 i2->h3
 5.03  -0.05  -0.14
 b->o h1->o h2->o h3->o
-1.02  6.27 -1.01 -5.07

[[2]]
a 2-3-1 network with 13 weights
options were - entropy fitting  decay=0.01
 b->h1 i1->h1 i2->h1
 -7.50  -1.09   3.76
 b->h2 i1->h2 i2->h2
 -7.03  -0.82   3.21
 b->h3 i1->h3 i2->h3
 6.54   0.56  -1.97
 b->o h1->o h2->o h3->o
 0.24 -4.98  7.82 -5.81

You can see that the weights change from step to step as expected.
How to retrieve the
"best" depends on your definition of best and the actual kind of CV
you did, but all the information is in there.


On Tue, Nov 16, 2010 at 3:30 AM,  <jrwang at itri.org.tw> wrote:
> Hi, I have a question regarding the MLearn in MLinterface package.  Assume I want to use nnet to build a nerual network prediction model.  I can observe performace of various parameters in nnet using MLearn to train and cross validate with my data.  Now, I want to retrive the model with best performance from my MLearn result.  What should I do?  I look the structure of returning result.  It seems a RObject inside the MLearn returing object containing a neural model.  Is this the best one?   If I can nnot do this from MLearn, is there any other method or package to do this?  Thanks,
>
> Best regards,
>
> Weihsin Wang, Ph.D.
> Bioinformatics Core Lab.,
> Biomedical Engineering Research Lab.,
> Industrial Technology Research Institute
> TEL:886-3-5913689
> FAX: 886-3-5820445
>
>
> ====================================================================
> 本信件可能包含工研院機密資訊，非指定之收件者，請勿使用或揭露本信件內容，並請銷毀此信件。
> This email may contain confidential information. Please do not use or disclose it in any way and delete it if you are not the intended recipient.
>        [[alternative HTML version deleted]]
>
>
> _______________________________________________
> Bioconductor mailing list
> Bioconductor at stat.math.ethz.ch
> https://stat.ethz.ch/mailman/listinfo/bioconductor
> Search the archives: http://news.gmane.org/gmane.science.biology.informatics.conductor
>