public double[] CreateInput(NeuralAgent agent, CardinalMovement action)
{
var input = InputCoder.Encode(agent);
InputCoder.EncodeAction(input, action);
return(input);
}
public void EncodeState(double[] input, NeuralAgent agent)
{
var space = agent.Anchor.Field.Space;
int px = agent.Anchor.Field.X - ViewRadius;
int py = agent.Anchor.Field.Y - ViewRadius;
int i = ObstaclesOffset;
for (int y = 0; y < ViewLength; y++)
{
for (int x = 0; x < ViewLength; x++)
{
var field = space[px + x, py + y];
input[i++] = field.IsOutside || field.HasObstacle ? 1 : 0;
}
}
px = agent.Anchor.Field.X - ViewRadius;
py = agent.Anchor.Field.Y - ViewRadius;
i = AgentsOffset;
for (int y = 0; y < ViewLength; y++)
{
for (int x = 0; x < ViewLength; x++)
{
var field = space[px + x, py + y];
input[i++] = field.HasAgent ? 1 : 0;
}
}
}
public double[] Encode(NeuralAgent agent)
{
double[] input = new double[EncodedSize];
EncodeState(input, agent);
EncodeGoal(input, agent);
return(input);
}
public void EncodeGoal(double[] input, NeuralAgent agent)
{
int dx = agent.Goal.Position.X - agent.Anchor.Field.X;
int dy = agent.Goal.Position.Y - agent.Anchor.Field.Y;
if (dx == 0 && dy == 0)
{
input[DirectionOffset] = 0;
input[DirectionOffset + 1] = 0;
}
else
{
var dn = Vector2.Normalize(new Vector2(dx, dy));
input[DirectionOffset] = dn.X;
input[DirectionOffset + 1] = dn.Y;
}
//var v1 = Vector2.Normalize(agent.Direction);
//var v2 = Vector2.Normalize(agent.Goal - agent.Position);
//var sin = Vector2.Dot(v2, v1);
//var cos = v1.CrossProduct(v2);
}
public AgentNetworkPrediction Predict(NeuralAgent agent)
{
return(Predict(InputCoder.Encode(agent)));
}