STUDENT ID NUMBERS and Names
21579617 - Mark Rogers
20206553 - James Long
22270627 - Siqi Wu


==== FOLDER DESCRIPTIONS
=== 1) code - all code written for this project
=== 2) data - data files used by code
=== 3) outline - the outline we submitted as part of hw2
=== 4) papers - a few papers we read for this project (just .pdfs)
=== 5) report - .tex and some .pdf for final report to be submitted
       	 - final report is in report/ms.pdf     
=== 6) results - output of code moved / stored here



============
============ MORE ON THE CODE FOLDER
============

==== contains code for fitting path algorithm for interval SVM
==== see folder: code/svmIntervalPath
Matlab code: the svmIntervalPath package
svmInterval.m: matlab function to fit an interval SVM by calling cvx functions.
svmIntervalPath.m: matlab function to compute the entire regularization path.
svmIntervalInit.m: matlab function to compute the initialization configurations.
svmIntervalFindNext0.m: matlab function to find the initial break point lambda.
svmIntervalFindNext.m: matlab function to iteratively find the next events.
vline.m: a convenient matlab function to plot verticle lines.
=== see folder ExamplePathPlots
ExamplePathPlots contains three plots that are relevant in our simulated example in the report: alphaPath.pdf, beta0Path.pdf and betaPath.pdf




==== Files for implementing ordinary "brute force" version of interval SVM

CODE FOR CROSS VALIDATION FRAMEWORK:
svm.m:  Takes as input a training set of interval data, training labels, a testing set of interval data, testing labels, and fixed hyperparameters lambda and rho and outputs the classification error and the model parameters beta and beta0. 

xval.m:  Partitions data into disjoint blocks and performs cross-validation to compute an estimate of the expected loss of the SVM with designated hyperparameters and interval dataset.

computeResults.m:  Takes as input an interval dataset and produces a file for each designated (lambda,rho) hyperparameter pair.

batch.m:  Script that calls driver function computeResults.m with indicate dataset and hyperparameters.



CODE FOR SUMMARIZING RESULTS:
collectResults.m:  Aggregates file outputs from computeResults.m to yield two matrices loss_values and L0_norm_b, which summarize loss values and L0 norm values as a function of the hyperparameters lambda and rho.  Also produces plots of all results.

collect.m:  Script that calls driver function collectResults.m with indicated directory of files produced by computeResults.m

analyze_features.m:  Takes as input summary file and outputs a text file containing features paired with the absolute values of their weights.  This is used to determine which features are important.

features.m:  Script that calls driver function analyze_features.m with indicated summary results filename




HELPER FUNCTIONS:
getParams.m:  Takes as input vectors of hyperparameters Lambda, Rho, and C and produces cell array Params, where Params{i,j} = [Lambda(i) Rho(j)]

quantile_ranking.m:  Normalizes data via the quantile ranking method

preprocess_data.m:  Removes zeros that are either all zeros or contain NaN's and then applies quantile ranking

mytiedrank.m:  Helper function to quantile_ranking.m which simply ranks elements of a vector and assigns equal ranks in the event of ties