==================================
     Topic Classifier
==================================

This is sample code that implements a nearest centroid classifier 
The classifier is written following the sklearn paradigm. 

Training data is assumed to be text documents that are made available
as dictionary objects with attributes: tf, df & ndocs where
tf = frequency of term in document
df = # docs the term in occurs in entire corpus
ndocs = total size of document corpus
(Assumption is that these documents are indexed in solr so that the 
term vectors are available easily)
The training data is a set of class/topic labels and associated exemplar
documents. 

SolrVectorizer is the estimator used to first transform
the exemplar document vectors to a sparse matrix of tfidf features
TopicClassifier is the core classifier implementation: it fits the 
training data by computing a centroid of the feature vectors. 
Calling the predict method on the TopicClassifier with test documents 
results in a topic label with the nearest centroid. 

Example runner script is test_core.py

(Note: you will need numpy, scipy and sklearn libraries installed
 in addition to python. Tested with numpy 1.7.1, scipy 0.12.0, 
sklearn 0.10, python 2.6.8)