DL Extension and Multiple Embedding Layers/Inputs
Guru
I am really starting to sink my teeth into the Deep Learning extension, and based on the last review, a lot of work has been done, so hat tip to the team internally!
In order to get acclimated to extension, I have been working through some past ideas that I have completed in Keras/TF2. One model was to jointly learn embeddings and concatenate that layer with additional inputs about the observation in order to fit a binary classifier. An example of that model is shown below.
Is it possible to generate a similar architecture in RM? From looking at the docs and the Tutorial processes, I see that it requires that the Embedding layer is the first layer (which of course makes sense), but its not clear how I could include additional attributes as inputs to a modelandselect the input that should feed the Embedding layer.
Also, it's not clear to me if we can perform other layer operations such as Flatten or Concatenate with the current version of the Extension.
In order to get acclimated to extension, I have been working through some past ideas that I have completed in Keras/TF2. One model was to jointly learn embeddings and concatenate that layer with additional inputs about the observation in order to fit a binary classifier. An example of that model is shown below.
Is it possible to generate a similar architecture in RM? From looking at the docs and the Tutorial processes, I see that it requires that the Embedding layer is the first layer (which of course makes sense), but its not clear how I could include additional attributes as inputs to a modelandselect the input that should feed the Embedding layer.
Also, it's not clear to me if we can perform other layer operations such as Flatten or Concatenate with the current version of the Extension.
0
