| """ |
| Author: Mélanie Gaillochet |
| Date: 2022-02-209 |
| |
| """ |
| from comet_ml import Experiment |
| import os |
| import numpy as np |
| import pandas as pd |
| import matplotlib.pyplot as plt |
|
|
| import torch |
|
|
| from Utils.sampler_utils import get_uncertainty_multiple_preds |
|
|
|
|
| class BaseMultiplePredsSampler: |
| """ |
| Base sampler for based on uncertainty from different predicted results |
| """ |
|
|
| def __init__(self, budget): |
| self.budget = budget |
|
|
| def base_sample(self, model, unlabeled_dataloader, device, sampling_type, transformation_type, num_dropout_inference=None, alpha_jsd=0.5, |
| data_aug_gaussian_mean=0, data_aug_gaussian_std=0): |
| |
| indice_list, mean_variance_list, max_variance_list, mean_jsd_list = get_uncertainty_multiple_preds( |
| model, unlabeled_dataloader, device, transformation_type, num_dropout_inference, alpha_jsd, data_aug_gaussian_mean, data_aug_gaussian_std) |
| |
| if 'MeanVariance' in sampling_type: |
| uncertainty = mean_variance_list |
| elif 'MaxVariance' in sampling_type: |
| uncertainty = max_variance_list |
| elif 'MeanJSD' in sampling_type: |
| uncertainty = mean_jsd_list |
|
|
| |
| arg = np.argsort(uncertainty) |
| querry_pool_indices = list(torch.tensor(indice_list)[arg][-self.budget:].numpy()) |
| uncertainty_values = list(torch.tensor(uncertainty)[arg][-self.budget:].numpy()) |
|
|
| return querry_pool_indices, uncertainty_values |
|
|
|
|
