Problems creating a customized pdf; asymmetric gaussian

Question

In order to create a customized pdf in zfit; Skew normal distribution (asymmetric gaussian), how can be called the error function? Seems that math.erf(x) and scipy.special.erf(x) do not works (they are tensor-like ones).

This is my code:

class AsymmetricGauss(zfit.pdf.ZPDF):
    _N_OBS = 1  # dimension, can be omitted
    _PARAMS = ['mean', 'std', 'alpha']  # the name of the parameters

    def _unnormalized_pdf(self, x):
        x = z.unstack_x(x)
        mean = self.params['mean']
        std = self.params['std']
        alpha = self.params['alpha']

        t = (x - mean)/std
        normal = (1/std*(math.sqrt(2*math.pi)))*z.exp(-(t)**2)
        cumulative = (1/2)*(1 + math.erf(alpha*t))
        return 2*normal*cumulative

This code raises error in the line:

cumulative = (1/2)(1 + math.erf(alphat))

. The error message is:

"NotImplementedError: Cannot convert a symbolic tf.Tensor (mul_2:0) to a numpy array. This error may indicate that you're trying to pass a Tensor to a NumPy call, which is not supported."

How can I solve this problem? Is there another better way to create a Skew normal distribution?

Answer 1

The answer is use tensorflow.math.erf(alpha*t) instead.