Currently we have param_search.py to run experiments with in-main-hardcoded logic. We need some interface to run experiment from code. I propose to add some class or function to train models in specified directory with dataset, props, lst, hook, number of seeds and unlabeled docs provided.
Then this code can be employed by param_search.py
Also we need public classes for datasets: CVDataset and HoldoutDataset.
Currently we have
param_search.pyto run experiments with in-main-hardcoded logic. We need some interface to run experiment from code. I propose to add some class or function to train models in specified directory with dataset, props, lst, hook, number of seeds and unlabeled docs provided. Then this code can be employed byparam_search.pyAlso we need public classes for datasets:CVDatasetandHoldoutDataset.