Thanks Mr. Ellison,
Your remark helped solve my error table problem.
However, I found a new one.
Now that I have my error tables, i realised that it is no good statistical
practise to validate a model, based one one error table.
So i should use a tool like K-fold CV.
ex: binary_model <- glm (y_binary~ x_value, family =
binomial,data=dataset)
cv.binary(binary_model,rand=NULL, nfolds=1000, print.details=TRUE)
This is no problem for the binary model, for the odds model this is not
the case.
Do you know a tool that can do this, or perhapes a way to implement it in
a monte carlo simulation?
(i added the way i solved the error table problem below)
Kind regards,
Tom.
>ERROR TABLE DILEMMA
>For a binary model there is no problem (here y has an outcome of 0 or 1)
>
>ex: pred_binary_model=(expit(predict(binary_model,tsample))>P)
> table_binary_model=table(pred_binary_model,tsample[,2])
>
TER_binary_model=sum(diag(table_binary_model[,]))/sum(table_binary_model)
>
> (error table1)
> pred_binary_model 0 1
> FALSE 28 95
> TRUE 4 114
> [1] 0.5892116 --> of correct classified cases
>
>Here there are 28 + 114 correctly predicted test cases, this results in
58.9% correct classified cases.
>A few more misclassified cases does not result in big departures from
this 58.9%.
>
>When i preform this on categorical data, i have to use frequency tables.
>This predicts the number of successes and the number of failures, per
interval.(odds per interval)
>So the error table does contain an outcome of odds for every given
interval.
>
>ex: (error table2)
> oddsPt
> pred_percent_model 0.00 0.16 0.37 0.84
> 0.05 1 0 0 0
> 0.16 0 1 0 0
> 0.34 0 0 1 0
> 0.78 0 0 0 1
> [1] 1 --> of correct classified cases
>
>As you can see, one misclassification will take disastrous proportions.
(~25% difference)
>The output of error table2 is interpretable, but it is not ideal, and
oversensitive to misclassification.
>
>I was able to solve this later problem by extracting the model
coefficients, and then using them in a function.
>Based on this function, i was able to write an error table equal to table
1.
Disclaimer: click here
[[alternative HTML version deleted]]
______________________________________________
[email protected] mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.
