KL Divergence Silently Skips Events

[jnoecker]

Events that appear in training but not test are probably useful evidence against authorship. JGAAP might consider using epsilon smoothing instead of dropping.