This pertains to Issue #1462 and #1513.
As of v0.9.5, LabelModel training assumes that LFs are conditionally independent given Y.
Are there any plans to update the loss function to reflect the algorithm described in multi-task weak supervision paper?
