Open Antinomies-O-S opened 8 months ago
To obtain enhanced samples, we performed downsampling on time-series data and introduced random noise. Regarding downsampling, if the original time-series data consists of 100 time points, we extract 50 points at regular intervals. For adding random noise, we use PyTorch's random number generation function to generate noise that aligns with the dimensions of the time-series data. Then, the augmented sample is utilized through a contrastive learning approach. This process helps the shared branch of the model become insensitive to both frequency and noise. Our paper is under review, and you can see more details when our paper is published.
您好,想要学习一下您的代码,但是关于数据结构那块我没有弄明白,您这边是对原始数据做了怎么样的处理然后再将它喂给神经网络的,能否给一个例子,多谢