From Cohen Courses
Jump to navigationJump to search

Perplexity is a measure to evaluate the predictive power of a probabilistic model. Lower perplexity means that the model is assigning higher likelihood to the unseen held out data. The parameters of the model is first estimated over the training data. Perplexity is defined as the exponential of the negative normalized likelihood over the held-out unseen data under the model in the question.