Projects

documentclassifier

A Java desktop application (J2SE 5+Swing Application Framework) for automatic classification of documents against a given training set. It has been developed, and is packaged, as a Netbeans project. It uses the stemmers created with Snowball (http://snowball.tartarus.org, released under the BSD ... [More] license) for text pre-processing, TF-IDF or the Bhattacharrya distance to rank the documents of the training set to the query document, and the K-NN algorithm to classify it. As of now, it only supports the classification of news from the ANSA website (http://www.ansa.it - The Italian main news agency), but the program has a modular architecture, that allows it to be extended by writing plugins for scraping the content of other websites, or other types of documents (PDF, DOC, ODT, etc...). [Less]

cell-knn

The goal of this project is to implement an efficient implementation of the k-NN algorithm on IBM's Cell processor. The k-nearest neighbor algorithm (k-NN) is a simple algorithm for machine learning. The algorithm projects objects to a multidimensional feature space. It then classifies objects ... [More] based on the closest (according to a distance metric) training examples in the feature space. An object is classified by a majority vote of its neighbors, with the object being assigned to the class most common amongst its k nearest neighbors. To evaluate the performance of our implementation(s) we classify elements in the MNIST database, which consists of up to 60000 images of handwritten digits. [Less]