Hi All,
Does anyone know if there is a way to build a DL4J neural network model to train the same model on multiple targets at the same time?
I’ve got a classification problem with multiple outputs but using the DL4J Feedforward Learner the only option seems to be to specify a single target. Is there a way around this? It is important for me that the network learns features in the hidden layers based on all targets rather than building a separate model for each target…?
unfortunately, that isn’t possible with our DL4J integration.
However, if you are willing to switch to our KNIME Deep Learning - Keras Integration, this kind of task is easy to accomplish, as you can have multiple output layers for a single network via the Keras Collect Layer node.
Many thanks for your quick reply. I used DL4J because I’m new to deep learning and it had is seemed to be a good starting point since it had simple example models on the example server that were for general multivariate datasets that are not related to for images or text.
I’l have a go at replicating the Dl4J using Keras - thanks for pointing me in the right direction!
you are most welcome
To get started with Keras on multivariate data, you might find Keras Autoencoder for Fraud Detection Deployment by Maarit informative.
In your case, the network will, of course, be different but the workflow showcases how to prepare your data for learning.
