The h2o package helps us train the neural networks. Deep learning models have many input parameters. In this exercise, the following parameters will be tested:
hyper_params <- list( hidden=list(c(5),c(80,80,80),c(75,75)), input_dropout_ratio=c(0.05,0.1,0.15,0.2,0.25), rate=c(0.01,0.02,0.10))
Here, we will test three structures: first, a network containing only a hidden layer with 25 neurons, then a network with three hidden layers with 32 neurons in each layer, and finally, a two hidden layer network with 64 neurons in each layer.
A neural network learns, and neurons progressively specialize in values for specific variables. If neurons are too specialized in the training set, there is a high risk of overfitting. ...
