| """ |
| Author: Mélanie Gaillochet |
| Date: 2021-03-25 |
| """ |
| import os |
| import numpy as np |
| import torch |
| import torch.utils.data as data |
|
|
| from Utils.sampler_utils import InfiniteSubsetRandomSampler, SubsetSequentialSampler |
|
|
|
|
| def update_kwargs(kwargs, method, config, sampling_type, labeled_indices=None, unlabeled_indices=None, |
| unlabeled_dataloader=None, dataset=None, batch_size=None, seed=42): |
| """ |
| This function updates the variables/params needed according to the AL method used |
| :param kwargs: kwargs that we want to update |
| :param method: AL method used |
| :param config: config file used |
| :param labeled_indices: (needed for VAAL and IMSAT_detach_unsup) |
| :param unlabeled_indices: (needed for VAAL and IMSAT_detach_unsup) |
| :param unlabeled_dataloader: (needed for VAAL) |
| :param dataset: (possibly augmented) dataset |
| :param batch_size: (needed for VAAL, VAE and ACNN) |
| :param debug: (needed for VAAL and VAE and) |
| :param seed: (needed for VAAL, VAE and ACNN) |
| """ |
|
|
| if 'Coresets' in sampling_type: |
| finite_seq_labeled_sampler = SubsetSequentialSampler(labeled_indices) |
| kwargs['finite_labeled_dataloader'] = data.DataLoader(dataset, sampler=finite_seq_labeled_sampler, batch_size=1, |
| drop_last=False, num_workers=config['training']['num_workers'], |
| persistent_workers=True, |
| pin_memory=True) |
| |
| if 'SemiSupervised' in method: |
| extra_sampler = InfiniteSubsetRandomSampler(dataset, labeled_indices + unlabeled_indices, shuffle=True) |
| extra_dataloader = data.DataLoader(dataset, sampler=extra_sampler, |
| batch_size=batch_size, drop_last=False, pin_memory=True, |
| num_workers=config['training']['num_workers']) |
| kwargs['extra_dataloader'] = extra_dataloader |
|
|
| return kwargs |
|
|
