unreal.LearningAgentsNormalization

class unreal.LearningAgentsNormalization

Bases: EnumBase

Normalization options for neural network inputs & outputs.

C++ Source:

  • Plugin: LearningAgents

  • Module: LearningAgents

  • File: LearningAgentsNeuralNetwork.h

AUTO_PER_DIMENSION: LearningAgentsNormalization = Ellipsis

Uses dataset derived standard deviation and mean as scales and biases to normalize/denormalize network dataflow. If a feature has multiple dimensions, a scale and bias will be computed for each dimension. <IMPORTANT> Hand picked scales and offsets values will NOT be used if normalization is set to “AutoPerDimension.”

Type:

2

AUTO_SHARED: LearningAgentsNormalization = Ellipsis

Uses dataset derived standard deviation and mean as scales and biases to normalize/denormalize network dataflow. If a feature has multiple dimensions, a single scale and bias will be computed for that feature. <IMPORTANT> Hand picked scales and offsets values will NOT be used if normalization is set to “AutoShared.”

Type:

1

MANUAL: LearningAgentsNormalization = Ellipsis

Uses user specified scales and biases to normalize/denormalize network dataflow at the cpp schema layer. <IMPORTANT> Hand picked scales and offsets values will only be used if normalization is set to “Manual.”

Type:

0