Classifying noisy signal samples - 1D convolution in fastai?

I have thousands of samples of a yes/no signal (the ack signal of a wireless IOT device) that I want to classify to their respective yes no. Every sample consists of a few hundred data points measured over some milliseconds.

My idea is to use 1D convolutional networks for this. However, I am not sure how this can be done using the fastai library. Any tips?


You can try to setup your own network with 1D conv blocks or try to use the standard architectures which automatically scale to the input image sizes. However, you will need 3 channel images to work with the standard architectures, e.g. ResNets. If you don’t you can copy the channels to get 3 or add a layer to the network input that does that in forward pass.