I’ve commonly encounter programmers choosing inputs that led to very slow AI learning. In this tutorial, I will show you the most recommended inputs so that you can get the best out of this library.
Relative magnitude means that the magnitude is the result from subtracting / dividing between two magnitudes. For example:
local relativeMagnitude = magnitude1 - magnitude2
local distance = position1 - position2
local rotationDifference = rotation1 - rotation2
local rotationValueNeededToFaceTheEnemy = math.tanh(distanceX / distanceZ) - currentRotationY
local healthChangedAmount = maxHealth - currentHealth
local healthRatio = currentHealth / maxHealth
Meanwhile magnitude is the value from zero. For example:
local position = primaryPart.Position
local rotation = primaryPart.Rotation
These two values can effect our neural networks very differently. Below, I will describe what will happens if you choose one of these values.
Because the programmer handled the differences in advance, the neural network doesn’t need to try to find the solution for this.
This will lead to more faster learning and require smaller neural network structure.
Since this is a raw value, the neural network now has to learn a solution related to output and it might not be even correct.
This will lead to slower learning and require larger neural network structure.
Directional equivalents means that the values provides the same information in terms of direction. For example, let’s say if we have these three values:
These can be represented as:
local rotationY = math.tanh(distanceX / distanceZ)
Because rotationY contains those two values, distanceX and distanceZ can be removed. However, if you want the distance for rotationY, then you need to calculate the distance using both distance X and Z.
Using redundant values will cause the learning to slow down as the neural network will have to find the connection between these values that already exists.
So you have to keep these two in mind:
Once you master those two, you can start seeing your AI learn much more faster.
That’s all for today!