public void Learn()
{
List <FloatTensor> rewards_list = new List <FloatTensor>();
List <FloatTensor> losses_list = new List <FloatTensor>();
for (int i = 0; i < history.Count; i++)
{
if (history[i][1] != null)
{
rewards_list.Add(history[i][1]);
}
if (history[i][0] != null)
{
losses_list.Add(history[i][0]);
}
}
FloatTensor rewards = Functional.Concatenate(this.controller.floatTensorFactory, rewards_list, 0);
FloatTensor losses = Functional.Concatenate(this.controller.floatTensorFactory, losses_list, 0);
var norm_rewards = rewards.Sub(rewards.Mean()).Div(rewards.Std().Add(0.000001f));
norm_rewards.Autograd = true;
var policy_loss = norm_rewards.Mul(losses.Neg()).Sum(0);
policy_loss.Backward();
optimizer.Step(rewards.Shape[0], 0);
history = new List <FloatTensor[]>();
}
public void Neg()
{
float[] data1 = { -1, 0, 1, float.MaxValue, float.MinValue };
int[] shape1 = { 5 };
var tensor1 = new FloatTensor(data1, shape1);
float[] data2 = { 1, 0, -1, -float.MaxValue, -float.MinValue };
int[] shape2 = { 5 };
var tensorNeg = new FloatTensor(data2, shape2);
tensor1.Neg();
for (int i = 0; i < tensor1.Size; i++)
{
Assert.AreEqual(tensor1.Data[i], tensorNeg.Data[i]);
}
}
public string processMessage(string json_message)
{
//Debug.LogFormat("<color=green>SyftController.processMessage {0}</color>", json_message);
Command msgObj = JsonUtility.FromJson <Command>(json_message);
if (msgObj.functionCall == "createTensor")
{
FloatTensor tensor = new FloatTensor(msgObj.data, msgObj.shape);
tensor.Shader = shader;
tensors.Add(tensor.Id, tensor);
Debug.LogFormat("<color=magenta>createTensor:</color> {0}", string.Join(", ", tensor.Data));
string id = tensor.Id.ToString();
return(id);
}
else
{
if (msgObj.objectType == "tensor")
{
//Below check needs additions/fix.
bool success = true;
if (msgObj.objectIndex > FloatTensor.CreatedObjectCount)
{
return("Invalid objectIndex: " + msgObj.objectIndex);
}
FloatTensor tensor = tensors[msgObj.objectIndex];
if (msgObj.functionCall == "init_add_matrix_multiply")
{
FloatTensor tensor_1 = tensors [msgObj.tensorIndexParams [0]];
tensor.ElementwiseMultiplication(tensor_1);
}
else if (msgObj.functionCall == "inline_elementwise_subtract")
{
FloatTensor tensor_1 = tensors [msgObj.tensorIndexParams [0]];
tensor.ElementwiseSubtract(tensor_1);
}
else if (msgObj.functionCall == "multiply_derivative")
{
FloatTensor tensor_1 = tensors [msgObj.tensorIndexParams [0]];
tensor.MultiplyDerivative(tensor_1);
}
else if (msgObj.functionCall == "add_matrix_multiply")
{
FloatTensor tensor_1 = tensors [msgObj.tensorIndexParams [0]];
FloatTensor tensor_2 = tensors [msgObj.tensorIndexParams [1]];
tensor.AddMatrixMultiply(tensor_1, tensor_2);
}
else if (msgObj.functionCall == "print")
{
return(tensor.Print());
}
else if (msgObj.functionCall == "abs")
{
// calls the function on our tensor object
tensor.Abs();
}
else if (msgObj.functionCall == "neg")
{
tensor.Neg();
}
else if (msgObj.functionCall == "add")
{
FloatTensor tensor_1 = tensors [msgObj.tensorIndexParams [0]];
FloatTensor output = tensor_1.Add(tensor_1);
tensors.Add(output.Id, output);
string id = output.Id.ToString();
return(id);
}
else if (msgObj.functionCall == "scalar_multiply")
{
//get the scalar, cast it and multiply
tensor.ScalarMultiplication((float)msgObj.tensorIndexParams[0]);
}
else
{
success = false;
}
if (success)
{
return(msgObj.functionCall + ": OK");
}
}
}
return("SyftController.processMessage: Command not found.");
}
