* Drop pandas requirement, switch cov matrix to numpy masked array * Implement sparse matrix support to reduce memory requirements on very large datasets * Changes suggested by petrelharp * Unit testing * Travis CI support * Better integration into pypi * requirements.txt