gtnn.network¶

gtnn.network.activation¶

Module with some basic neuron activation functions.

class gtnn.network.activation.Identity[source]¶

Identity activation function

derivative(x)[source]¶

Returns:	always 1.0

value(x)[source]¶

Returns:	x

class gtnn.network.activation.LogSigmoid(outMin, outMax)[source]¶

Scaled sigmoid activation function

derivative(x)[source]¶

Returns:	scaled sigmoid’s derivative

value(x)[source]¶

Returns:	scaled sigmoid

gtnn.network.net¶

class gtnn.network.net.Net(nInput, nOutput, graph=<Graph object, directed, with 0 vertices and 0 edges at 0x2ab54322ce10>)[source]¶

Bases: object

This is a wrapper for the graph_tool.Graph object. One can register vertex/edge properties and use them in algorithm implementation. It also implements basic forward and backward operations.

addEdgeProperty(name, typeName)[source]¶

addVertexProperty(name, typeName)[source]¶

backward(outputErr=[])[source]¶

forward(inputVals=[])[source]¶

prepare()[source]¶: Computes ans saves the topological sort of the graph for future use.