[dattri/examples] add data cleaning example

[TheaperDeng]

Description

1. Motivation and Context

Add an example for data selection.

2. Summary of the change

Add an example for data selection. This script find the top-K low-quality data that "harm" the validation performance the most. The script also evaluate the test performance after removing the top-K low-quality data points.

Namespace(method='explicit', device='cuda', seed=0, train_size=10000, val_size=10000, test_size=5000, remove_number=100)
Test loss: 0.39373043179512024                                                                                                                                                        
Accuracy: 88.52                                                                                                                                                                       
Test loss (after): 0.38540247082710266
Accuracy  (after): 88.94

3. What tests have been added/updated for the change?

• Application test: If you wrote an example for the toolkit, this test should be added.

[tingwl0122]

LGTM

[tingwl0122]

LGTM