Refer to the course intro handout for guidelines on working with other students. Remember to submit your solutions in electronic form to cis830ta@ringil.cis.ksu.edu and produce them only from your personal data, source code, and notes (not common work or sources other than the codes specified in this machine problem). If you intend to use other references (e.g., codes downloaded from the CMU archive, NRL archive, or other software repositories such as those referenced from KD Nuggets or the instructors "related links" page), get the instructor’s permission, and cite your reference properly.

1. (40 points) Running ID3. For this machine problem, you will use your course accounts on the KSU CIS department KDD cluster. Note: Your accounts may not yet be operational by the time this assignment is distributed – check the course web page for the latest information.

1. Log into your course accounts on Topeka and Salina (when they are ready) and download the files:

http://ringil.cis.ksu.edu/Courses/Spring-2000/CIS830/Homework/Problems/HW2/MLC++-2.01.tar.gz
http://ringil.cis.ksu.edu/Courses/Spring-2000/CIS830/Homework/Problems/HW2/db.tar.gz
This is the pre-compiled binary for RedHat Linux 6.x.
2. Follow the instructions in the MLC++ manual (Utilities 2.0, in your first notes packet) for installing it in your scratch directory on Topeka and Salina:
/cis/{topeka | salina}/scratch/CIS830/yourlogin
3. Follow the instructions in the MLC++ tutorial (also in your first notes packet) to run the ID3 inducer on the following data sets from the UCI Machine Learning Database Repository: Breast, Tic-Tac-Toe, Ionosphere. Use the .test files for testing. Turn in a PostScript file containing the decision tree and another file containing a table of training and test set accuracy values for each data set.

2. (30 points) Using NeuroSolutions.

1. Download the NeuroSolutions 3.02 demo from http://www.nd.com and install it on a Windows 95, 98, NT 4.0, or NT 5 (Windows 2000) machine.
2. Use the NeuralWizard (which is fully documented in the online help for NeuroSolutions 3) to build a multilayer perceptron for learning the second sleep stage data set. Your training data file should be sleep2.asc and your desired response file should be sleep2t.asc. Use a 20% holdout data set for cross validation. Report both training and cross validation performance (mean-squared error) by stamping a MatrixViewer probe on top of the (octagonal) costTransmitter module and recording the final value after training (for the default of 1000 epochs).

3. (30 points) Using Hugin.
Download the Hugin Lite demo from http://www.hugin.com and use it to build the full Bayesian network for the Forest Fire example from Lecture 19, using your own subjective estimates of CPTs. Make sure that all your probability values are legitimate (specifically, that they have the proper range and marginalize properly). Turn in, by e-mail, a screen shot of your BBN and attach it as a Hugin file titled ForestFire.hkb.

Extra credit (15 points): Learning time series data with NeuroSolutions.
For all parts, turn in training (80%) and cross validation (20%) error values.

1. Train a Jordan-Elman network for the same task and report the results. Use the default settings and the input recurrent network (the upper left entry among the 4 choices. Take a screen shot of your artificial neural network after training (in Windows, hit Print-Screen and paste the Clipboard into your word processor).
2. Train a time-delay neural network for the same task and report the results.
3. Train a Gamma memory for the same task and report the results.

11. The Lumiere Project: Bayesian User Modeling for Inferring the Goals and Needs of Software Users (Horvitz, Breese, Heckerman, Hovel, and Rommelse)
12. Symbolic Causal Networks for Reasoning about Actions and Plans (Darwiche and Pearl)
13. KDD for Science Data Analysis: Issues and Examples (Fayyad, Haussler, and Stolorz)