Is there a 'Pixel classification layer' equivalent for 1 dimensional vector 'Deep Network Designer'?

I am trying to test a theory where my input is a 1-dimensional vector with N elements. I wanto to use an 'Auto-encoder' like network structure to compute a new N element vector which should ideally match the output sequence. Also, each one of the N outputs can belong to one of three classes. Is there a way to do this?

Answer 1

[Raunak Gupta]

Hi,

There is 1-D convolutional layer which you can build using the methods described here and here. This way a encoder and decoder network can be built in 1-D. After that you can attach pixel classification layer at the end to complete the workflow. Since the output size is calculated automatically by pixelClassificationlayer, the 1-D input size will also work (Make sure to give appropriate batch size too). Using 1-D convolutional layer described in above answers you can build the autoencoder model.