In this pr I have the results of running two generative models on a 50k training set. The first model assumes each label function is conditionally independent and the second model takes correlation structure into account (dependency aware). It turns out that modeling the correlation structure is better for performance than assuming independence. Let me know what you think.
In this pr I have the results of running two generative models on a 50k training set. The first model assumes each label function is conditionally independent and the second model takes correlation structure into account (dependency aware). It turns out that modeling the correlation structure is better for performance than assuming independence. Let me know what you think.