DataTransform_FD funtion implement is inconsistent with the paper? about tfc-pretraining HOT 1 CLOSED

Hi,

Thanks for pointing that out. The four lines you mentioned are the old version of frequency augmentation, they should be commented out. The augmentations are implemented by the following:

def DataTransform_FD(sample, config):
    aug_1 = remove_frequency(sample, pertub_ratio=0.1)
    aug_2 = add_frequency(sample, pertub_ratio=0.1)
    aug_F = aug_1 + aug_2
    return aug_F

def remove_frequency(x, pertub_ratio=0.0):
    mask = torch.FloatTensor(x.shape).uniform_() > pertub_ratio  # maskout_ratio are False
    mask = mask.to(x.device)
    return x*mask

def add_frequency(x, pertub_ratio=0.0):

    mask = torch.FloatTensor(x.shape).uniform_() > (1-pertub_ratio)  # only pertub_ratio of all values are True
    mask = mask.to(x.device)

Moreover, we have updated the TFC implementation. Please check more details in the Updates on Jan 2023 section of the repo readme. In summary:

• Fixed bugs, cleaned the codes, and added comments for better understanding.
• For the contrastive encoders (in both time and frequency domains), we replaced the 3 layers of CNN blocks with 2 layers of Transformer.
• For the downstream classifier, we added a KNN classifier in parallel with the original MLP (2-layer) classifier.
• Shared more ideas that may improve the TF-C framework in follow-up works.

from tfc-pretraining.