Activation Function
From Encog Machine Learning Framework
(Redirected from Activation function)
An activation function is a mathematical function used by a neural network to scale numbers to a specific range. There are several activation functions supported by Encog. Some activation functions have parameters that allow such attributes as the slope of the activation function to be defined.
Contents |
Most Common
- Sigmoid Activation Function
- Hyperbolic Tangent Activation Function
- Linear Activation Function
- Elliott Activation Function
More Activation Functions
- Gaussian Activation Function
- Logarithmic Activation Function
- Ramp Activation Function
- Sine Activation Function
- Step Activation Function
Specialty Activation Functions
Comparison of Activation Functions
| Encog Class | Range | PMML | Description |
|---|---|---|---|
| ActivationBiPolar | -1 or 1 | unavailable | |
| ActivationCompetitive | n/a | unavailable | |
| ActivationElliott | [0 to 1] | unavailable | |
| ActivationElliottSymmetric | [-1 to 1] | Elliott | |
| ActivationGaussian | [0 to 1] | Gauss | |
| ActivationLinear | [all real numbers] | identity | |
| ActivationLOG | [-1 to 1] | unavailable | |
| ActivationRamp | n/a | unavailable | |
| ActivationSigmoid | [0 to 1] | logistic | |
| ActivationSIN | [0 to 1] | Sine | |
| ActivationSoftMax | n/a | unavailable | |
| ActivationStep | n/a | threshold | |
| ActivationTANH | [-1 to 1] | tanh |
