Closed yzj19870824 closed 5 years ago
Hi @yzj19870824 — great question. You're right, the tutorial mainly highlights the MMTL functionality in the modeling pipeline, but this framework is agnostic to the way that we create/build training sets.
I'll also note that the best of MeTaL has been refactored to the v0.9 release of Snorkel — please see the repository of tutorials for examples of how weak supervision fits into the context of MMTL!
It's supervised learning In mmtl_basics.ipynb, how to use weak supervised learning with MMTL, just like multitask.ipynb does?