Hi, if i am correct, I found a critical problem on your code.
The problem is in the run_split.py -SplitMnistGenerator()
In the classes, it vertically stacks two labels of data without data shuffling. As a result, when we call coreset_method() in vcl.py- run_vcl(), it may sample only one label of data if the coreset size is small.
Hi, if i am correct, I found a critical problem on your code.
The problem is in the run_split.py -SplitMnistGenerator() In the classes, it vertically stacks two labels of data without data shuffling. As a result, when we call coreset_method() in vcl.py- run_vcl(), it may sample only one label of data if the coreset size is small.
Can you check this problem? Thanks