public Tensor categorical_crossentropy(Tensor target, Tensor output, bool from_logits = false)
{
// https://github.com/fchollet/keras/blob/f65a56fb65062c8d14d215c9f4b1015b97cc5bf3/keras/backend/cntk_backend.py#L1480
var _output = In(output);
var _target = In(target);
if (from_logits)
{
var result = C.CrossEntropyWithSoftmax(_output, _target);
// cntk's result shape is (batch, 1), while keras expect (batch, )
CNTK.Function r = C.Reshape(result, NDShape.CreateNDShape(new int[] { }));
return(Out(r));
}
else
{
// scale preds so that the class probas of each sample sum to 1
var o = C.ElementDivide(_output.function, C.ReduceSum(_output, Axis.EndStaticAxis()));
var eps = Constant.Scalar(epsilon(), DeviceDescriptor.CPUDevice);
var omeps = Constant.Scalar(1.0 - epsilon(), DeviceDescriptor.CPUDevice);
// avoid numerical instability with _EPSILON clipping
o = C.Clip(o, eps, omeps);
CNTK.Function r = C.Negate(C.ReduceSum(C.ElementTimes(_target, C.Log(_output)), Axis.EndStaticAxis()));
return(Out(r));
}
}
public Tensor binary_crossentropy(Tensor output, Tensor target, bool from_logits = false)
{
log(new { output, target, from_logits });
var _output = new Variable(In(output).function);
var _target = new Variable(In(target).function);
if (from_logits)
{
_output = C.Sigmoid(_output);
}
// scale preds so that the class probas of each sample sum to 1
var eps = InConstant(epsilon());
var omeps = InConstant(1.0);
// avoid numerical instability with _EPSILON clipping
_output = C.Clip(_output, eps, omeps);
var a = new Variable(C.Negate(C.ElementTimes(_target, C.Log(_output))));
var b = new Variable(C.Negate(
C.Minus(C.ElementTimes(InConstant(1.0), _target),
C.Minus(C.Log(InConstant(1.0)), _output))));
_output = a + b;
return(Out(_output));
}
public Tensor Log(Tensor x)
{
return(Out(C.Log(In(x))));
}