Re: Adjusting

From: Richard Ulrich (Rich.Ulrich_at_comcast.net)
Date: 02/28/05 

Date: Mon, 28 Feb 2005 17:45:35 -0500

On 27 Feb 2005 11:51:28 -0800, "Predictor" <predictr@bellatlantic.net>
wrote:

>
> Radford Neal wrote:
[snip, some about unequal sample sizes.
]
> >
> > One way is to just do weighted regression, with the ratio of weights
> > for majority and minority cases being proportional to the bias
> factor.
> >
> > Radford Neal
>
> I personally have no empirical or theoretical evidence for this, but
> this course has been suggested many times, both in print and on Usenet.
> Search Usenet for:

Not Usenet; search GROUPS with google, and in particular,
search comp.ai.neural-nets.

Greg Heath writes some good advice, as does Warren Sarle.
But many of the questions and answers seem to be
about <whatever> NN methods instead of being Logistic.

>
> Heath unbalanced
>
> or...
>
> "Don Cram" unbalanced
>
> These sources seem to indicate that models improve by stratified
> sampling of classes. Some suggest using all exemplars from the rare
> class, and rotating through exemplars from the majority class during
> iterations of the logistic regression fitting process.
>
> Presently, I am operating under the assumption that the balancing
> suggestion is appropriate, but am left with the question of how to
> adjust the logistic regression constant afterward.

I read a couple of threads. I did not find anything new,
in 20 minutes or so.

You can *sample* from the larger class to save computing time,
at the risk of losing data-generality. (I guessed that.)

"Classification of cases" is problematic... of course.
What do you do with those Bayesian-count probabilities?
How do you count errors? 

-- 
Rich Ulrich, wpilib@pitt.edu
http://www.pitt.edu/~wpilib/index.html

Relevant Pages

  • Re: Single-Factor-Cox-Regression
    ... allowed within Cox-Regression) ... Like the logistic regression, which it is sort-of an extension ... Cox regression models hazards and hazard ratios. ... Logistic regression is to require at least 20 more cases in the ...
    (sci.stat.math)
  • Re: Approximate solution to linear regression
    ... Construct an ensemble of regression models, ... these are binary variables. ... So my idea to use a logistic regression to classify 15% of the ...
    (sci.stat.consult)
  • Re: Logistic regression or Poisson regression (log linear)
    ... I've been looking at some analyses with similarly sparse data. ... prior to logistic regression will probably lose a lot of information. ... > treatment more often than with the experimental treatment. ...
    (sci.stat.consult)
  • Re: Regression Inference and Data Splitting
    ... The statement is: " Regression ... is logistic regression, it has relevance to all regression-type models. ... Internal validation of predictive models: ... Several internal validation methods are available that aim to ...
    (sci.stat.math)
  • Re: logistic regression
    ... > logistic regression? ... as log-- "Logistic regression" has easy ... a more subtle solution, ... diagnostics than what you can get on OLS regression. ...
    (sci.stat.math)