Project 1

Files

• writeup.pdf: PDF document which answers questions from the project requirements.
• common.py: Contains methods for generating random gaussian data sets and other common methods used throughout.
• svm.py: Contains SVM class implementation.
• svmDriver.py: Driver for the SVM. Used to call the SVM from command line.
• svmDriver.py: Driver for the Regression. Used to call the regressions from command line.
• regression.py: Contains the linear and logistic regression implementation.

Dependencies

• Python 3
• Python Packages
  • numpy
  • matplotlib
  • math
  • cvxpy

General Execution Notice

As with all scripts, depending on the operating system being used, execution may vary slightly. When running as a script directly (for example: svmDriver.py) it may be necessary to preface the script with a path (./svmDriver.py) or grant execution permission to the script (for example: chmod 777 svmDriver.py). To avoid this problem, I prefaced script executions with the appropriate Python version python3, for example python3 svmDriver.py.

SVM

Run Instructions

The SVM is implemented in svm.py, however, use svmDriver.py to run the SVM from command line.

For the default SVM (C=1, plot graphically, do not print report), execute python3 svmDriver.py. The optional flag --c allows a custom constant c value to be used, for example, python3 svmDriver --c=10. The optional flag --noPlot will stop the plot from appearing graphically, for example python3 svmDriver.py --noPlot. The optional flag --printReport will print basic SVM information (information requested in question 1b) to the console. The optional flag --brute should only be used if a report is being printed, to calculate leave-one-out error by brute force retraining (will greatly increase run-time).

For a full breakdown of the driver parameters, run svmDriver.py -h for help.

Execution Steps

These are the steps used to generate the data discussed in the write up.

1. python3 svmDriver.py --printReport --brute
   • This command will generate an SVM where C=1, print vital information to console, and calculate the leave-one-out error by brute-force retraining.
   • Output is Figure 1 in the report.
2. python3 svmDriver.py --printReport --c=0
3. python3 svmDriver.py --printReport --c=10
4. python3 svmDriver.py --printReport --c=100
5. python3 svmDriver.py --printReport --c=1000

Regression

Run Instructions

The regression is implemented in regression.py, however, use regressionDriver.py to run the regression from command line.

For the linear regression, execute python3 regressionDriver.py linear. For the logistic regression, execute python3 regressionDriver.py logistic.

For a full breakdown of the driver parameters, run regressionDriver.py -h for help.

Execution Steps

These are the steps used to generate the data discussed in the write up.

1. python3 regressionDriver.py linear --loo
2. python3 regressionDriver.py logistic --loo