[Validation] To compare `snorkel`'s weak supervised learning, `pulearn` could be used as a baseline

[Nikronic]

Well, it assumes we only have positive labels and unlabeled exists. In our case, we want to use this model to see how many of rejected cases this model gets right:

1. How many rejected case that we have their rejection label
2. How many rejected case that have weak labels in our dataset

Note: since pulrean is sklearn based library, it is easy to integrate and test, so does not need lots of SWE for integration and testing. I personally think I should just do it in a notebook for testing and if it had something worth mentioning, then it should be used as another print statement inside reporting metrics for a validation and nothing else.

Ref:

1. https://github.com/pulearn/pulearn
2. https://pulearn.github.io/pulearn/doc/pulearn/