This PR starts off post PSB rejection and the overall goal here is to update the snorkel package and other outdated packages.
Reasons for updating:
More stable Generative Model. The parameter the model estimates is mu that represents the likelihood of a label function emitting a label (abstain, negative, positive) given the true class.
Before Upgrade:
mu - [
[ 0.106, -0.0014],
[0.000, 0.000],
[0.000, -0.001]
]
After Upgrade:
mu - [
[1.0000e-07, 1.0000e-07],
[4.1568e-03, 4.5846e-03],
[1.0000e-07, 1.0000e-07]
]
^ floating point calculations better reflect probability estimates.
Able to obtain weights for each label function. This provides better insight to what label functions the model considers important
Code style has typing which makes reading/using their code easier
Overdue task since I have to retrace my steps on this project.
Adding @dhimmel to keep him up to date while he is afk during the next few days. (Another PR will come shortly)
This PR starts off post PSB rejection and the overall goal here is to update the snorkel package and other outdated packages.
Reasons for updating:
Adding @dhimmel to keep him up to date while he is afk during the next few days. (Another PR will come shortly)