Incorrect neural network schema in training output

Question

I'm training a model in Azure ML Studio and the Net# specification I'm using doesn't match the NET# specification in the training output.

Here's my experiment -

and here are my NN params -

and finally here is the NET# specification in the Hyperparams output -

It's not using two hidden layers and it's also using sigmoid instead of ReLu. Is this expected behavior?

Thanks!

Answer 1

There is an issue with using custom NET# and parameter sweeps together: it switches over to using the default fully connected topology.

Unfortunately, the workaround is to train the model for each parameter value separately.

-Roope