Training KPIS - m-marini/wheellyj GitHub Wiki

TDAgent

The TDAgent generates the following kpis at each step

Key	Values
s0.*	the network signals
actions.*	the actions
reward	the reward

At each training mini-batch step

Key	Values
delta	the delta value (error)
avgReward	the average reward
trainingLayers.*	the training layer values
trainedLayers.*	the trained layer values
actionMasks.*	the action masks
grads.*	the output gradients
deltaGrads.*	the output gradients multiply by delta

Batch trainer

The batch trainer generates the following kpis at each training mini-batch step

Key	Values
delta	the delta value (error)
avgReward	the average reward
trainingLayers.*	the training layer values
trainedLayers.*	the trained layer values
actionMasks.*	the action masks
grads.*	the output gradients
deltaGrads.*	the output gradients multiply by delta

---

Release 0.17.0