I have data of shape (N_samples,N_channels,N_features_per_channel)
I would like to implement a layer, where each channel is fully connected to a set of output nodes, and there is no weight sharing between the channels weights.
Can this achieve what I am looking for ?
nn.Conv1d(in_channels=N_channels, out_channels=N_channels_out, kernel_size=1, groups=N_out_features_per_channel)