|
SmartEngine
1.6.0
|
GeneticTrainer mutation info More...
#include <GeneticTrainer.h>
Public Attributes | |
| MutationTarget | target = MutationTarget::Auto |
| Specifies what weights to mutate. More... | |
| float | nodesToMutate = -1.0f |
| How many trainable nodes in the graph we should mutate More... | |
| float | targetsToMutate = 0.08f |
| How many weights / neurons per node to mutate. More... | |
| float | standardDeviation = 1.0f |
| A random value of this standard deviation will be added to each weight selected for mutation. More... | |
| float | minStandardDeviation = 1e-3f |
| The minimum standard deviation we are allowed to achieve when AdaptiveStandardDeviation is set to true. More... | |
| bool | adaptiveStandardDeviation = true |
| If set to true, the standard deviation will change with the learning rate. If the network isn't learning, the standard deviation will lower, allowing for fine grain tuning. If learning is happening quickly, the standard deviation will raise, allowing for larger jumps in progress. More... | |
| int | stepsUntilLowerStandardDeviation = 1 |
| The number of steps that the best network is in the top unchanged percent before we lower the mutation standard deviation. This is only used if AdaptiveStandardDeviation is set to true. More... | |
| int | stepsUntilRaiseStandardDeviation = 1 |
| The number of steps that the best network is not in the top unchanged percent before we raise the mutation standard deviation. This is only used if AdaptiveStandardDeviation is set to true. More... | |
GeneticTrainer mutation info
| bool SmartEngine::MutationInfo::adaptiveStandardDeviation = true |
If set to true, the standard deviation will change with the learning rate. If the network isn't learning, the standard deviation will lower, allowing for fine grain tuning. If learning is happening quickly, the standard deviation will raise, allowing for larger jumps in progress.
| float SmartEngine::MutationInfo::minStandardDeviation = 1e-3f |
The minimum standard deviation we are allowed to achieve when AdaptiveStandardDeviation is set to true.
| float SmartEngine::MutationInfo::nodesToMutate = -1.0f |
How many trainable nodes in the graph we should mutate
A value [0..1] will mutate a percent of all nodes.
A value of 0.0 exactly means mutate all nodes.
An negative integer (-infinity, 0) will mutate that exact number of nodes.
| float SmartEngine::MutationInfo::standardDeviation = 1.0f |
A random value of this standard deviation will be added to each weight selected for mutation.
| int SmartEngine::MutationInfo::stepsUntilLowerStandardDeviation = 1 |
The number of steps that the best network is in the top unchanged percent before we lower the mutation standard deviation. This is only used if AdaptiveStandardDeviation is set to true.
| int SmartEngine::MutationInfo::stepsUntilRaiseStandardDeviation = 1 |
The number of steps that the best network is not in the top unchanged percent before we raise the mutation standard deviation. This is only used if AdaptiveStandardDeviation is set to true.
| MutationTarget SmartEngine::MutationInfo::target = MutationTarget::Auto |
Specifies what weights to mutate.
| float SmartEngine::MutationInfo::targetsToMutate = 0.08f |
How many weights / neurons per node to mutate.
A value [0..1] will mutate a percent of all weights / neurons.
An negative integer (-infinity, 0) will mutate that exact number of weights / neurons. A negative integer can only be specified if targeting neurons instead of weights.
1.8.20