topo.in

This input file defines the topology of the network. Here is an example, taken from the kappa problem:

3 number_of_grids
1 10 10 size_of_input_hidden_output_grid
1 flag_bias_to_output_connections
0 flag_direct_connections
0 flag_load_net
0.25 0.0 wire_sigma_init
1 1.0 1.0 flag_pao___steep_bias

Row 1
Number of layers in the network, set to 3 for the input, first hidden and kernel layers. Adjacent layers all all-to-all connected. The nodes in the first hidden layer are connected to the bias node.

Row 2
Three integers indicating the numbers of nodes in the input, first hidden and kernel layers (in this order).

Row 3
Flag variable for bias-to-kernel connections. If set to 1, the Gaussian kernels are connected to the bias node. If set to 0, these connections are omitted.

Row 4
Flag variable for direct connections. If set to 1, the input layer is connected to the kernel layer. If set to 0, direct connections are omitted.

Row 5
Flag variable for loading a network into internal memory. If set to 0, no network is loaded, and the weights are randomly initialised. If set to 1, the network parameters are read in from file w_0.net.

Row 6
First number: Standard deviation of the Gaussian distribution from which the adaptable weights are drawn during initialisation. Leave the second number unchanged.

Row 7
First number: Flag variable for RVFL. If set to 1, an RVFL network is created, in which the input-to-first-hidden-layer weights and the bias-to-first-hidden-layer weights are held constant. The former are drawn from a Gaussian with a standard deviation given by the second number. The latter are drawn from a Gaussian with a standard deviation given by the third number. If the first number is set to 0, a standard Gaussian mixture network is created, in which all parameters are adapted.


Last modified: 19 May 2000