Why is there only a frft pooling layer in the code? Can you provide code related to time series prediction tasks? I would like to use frft as a feature extraction layer. Thank you
Sorry to bother you again. I want to know about an issue. Your trainable FRFT (Fractional Fourier Transform) indeed performs well, but on a relatively small dataset, it doesn't perform as well as the ordinary FFT (Fast Fourier Transform). I want to understand if this is because the dataset is too small, leading to insufficient learning of the parameters in the FRFT, and thus causing its performance to be inferior to that of the ordinary FFT.Thank you.
Now there is a signal classification task,the input signal is a one-dimensional real number signal, that is, a one-dimensional sequence. I use trainable FRFT to process the input signal, and then use CNN and linear layers to classify the processed data. Here, the CNN input is dual channel, and the transformed real and imaginary parts are processed separately. May I ask why the effect is worse than using only CNN and linear layers, and where is the error? Thank you for your answer。By the way, the code I referenced is GitHub page of torch-frft,only added a CNN and linear layer at the end