Neural Network Factory
Neural network factory provide convenient helper functions to define common neural networks.
MXNet.mx.MLP — Method.
MLP(input, spec; hidden_activation = :relu, prefix)
Construct a multi-layer perceptron. A MLP is a multi-layer neural network with fully connected layers.
input::SymbolicNode: the input to the mlp.
spec: the mlp specification, a list of hidden dimensions. For example,
[128, (512, :sigmoid), 10]. The number in the list indicate the number of hidden units in each layer. A tuple could be used to specify the activation of each layer. Otherwise, the default activation will be used (except for the last layer).
hidden_activation::Symbol: keyword argument, default
:relu, indicating the default activation for hidden layers. The specification here could be overwritten by layer-wise specification in the
specargument. Also activation is not applied to the last, i.e. the prediction layer. See
Activationfor a list of supported activation types.
prefix: keyword argument, default
gensym(), used as the prefix to name the constructed layers.
Returns the constructed MLP.