SVM Classification

SVM classification with no augmentations, with clustering, and with dimensionality reduction. Also performed Naive Bayes analysis with harthur/classifier.

cluto:

CLMethod=RB. Computed clusters used as meta-features in document feature vectors.

svm_light:

Utilized default kernel, binary classifier for each cluster. "nc" denotes classification done without clustering metafeatures. test_documents_key corresponds to the correct classification for the test documents.

doc2mat-1.0:

tfidf.py- transforms a given matfile (generated from running doc2mat script) into .dat file.

Input arguments: [name of .mat file] [name of .dat file (optional)]

Clustering results found in svm_light

3c_, 4c_, 5c_- clustering done with 3, 4, and 5 clusters respectively, using TF-IDF of trigrams. Training documents put into groups based on clustering performed with only the training documents (as opposed to with the testing docs).

cte files

Classification performed with larger data set, using categorized_text_edited.txt. doc2mat_cte.py- converts categorized_text_edited.txt into docfile compatible with cluto and generates corresponding .rlabel file.

full_data_plots:

.ps files for cte_docfile