Homework 7. Stat 540 Spring 1999.

Objective Allow a remote user to perform prediction on a binary outcome variable, using a set of covariates. Use both logistic regression and a neural network for the prediction to allow a comparison of results.

Prediction should be done out of sample, that is on the validation dataset.

Make sure that your procedure works on the dataset found in
http://www-stat.wharton.upenn.edu/~waterman/DataSets/class_06.data

Deliverables (by 3/4/99):

An HTML interface page that allows a user to run both the logistic regression (additive at this stage) and neural network. It must have an input area in which the user identifies the URL of their dataset, and chooses the number of nodes for the net.
A cgi perl script that receives the request, fetches the data, runs the R process and returns the results. Results should be in the form of two tables for the out of sample predictions.
Thought question: consider how you might identify areas in the covariate space in which predictions differ between the two methods.

Additional comments added 2/27/99.
I've cleaned up the trial data set to be used for testing your code on. It's called "new.uva" and is a tab delimited text file. The URL is "http://www-stat.wharton.upenn.edu/~waterman/DataSets/new.uva"

Please put your interface page in
/home/username/public_html/Stat540s99/homework07/interface.html
and the Perl script in
/home/username/public_html/cgi-bin/homework07.pl

Remember to rescale, to (0,1), the age variable before you use it in the net. In general, you will need to be able to identify the continuous predictor variables so you can rescale them. Use the command is.real to do this.

Don't forget that I have put a link to the Neural nets help files. help pages.

As a matter of interest, here is the S-Plus code that I used to produce the page of neural net graphics. You can see how I constructed the perspective plots. Note that parts of this code are out of date now, as I had the uva dataset labeled differently when I wrote it, the file names differently and everything in different locations! (But it's still a useful example.)