Viewing model architecture

Google Cloud Build with AI Custom ML & MLOps
1

Hello! How can I interpret the feedforward NN model architecture described in AutoML logs (after training a model using AutoML in VertexAI)?

I understand the base structure described in https://cloud.google.com/automl-tables/docs/logging

But I am not sure this describes the full architecture. For example, if num_neurons = 256, and num_layers = 2, how do I know how many neurons on each layer? Or for dropout = 0.5, in which layer is the dropout happening?

Any sources your recommend that might explain this a bit better?

Thank you very much in advance for your help! I have been researching this and have found no clear explanation

2

Hi,

It seems the doc1 you shared is the only one I can find that describes the feedforward NN model architecture.

Keep in mind that AutoML Tables is covered by the Pre-GA Offerings Terms of the Google Cloud Terms of Service.

But I encourage you to file a feature request here2 to have the full model architecture details added to the log entries.