def _log_cdf(self, x):
loc = tf.convert_to_tensor(self.loc)
scale = tf.convert_to_tensor(self.scale)
safe_x = self._get_safe_input(x, loc=loc, scale=scale)
log_cdf = np.log(2 / np.pi) + tf.math.log(
tf.atan((safe_x - loc) / scale))
return tf.where(x < loc,
dtype_util.as_numpy_dtype(self.dtype)(-np.inf),
log_cdf)
def _log_cdf(self, x):
loc = tf.convert_to_tensor(self.loc)
scale = tf.convert_to_tensor(self.scale)
with tf.control_dependencies(self._maybe_assert_valid_sample(x, loc)):
safe_x = self._get_safe_input(x, loc=loc, scale=scale)
log_cdf = np.log(2 / np.pi) + tf.math.log(
tf.atan((safe_x - loc) / scale))
return tf.where(x < loc,
dtype_util.as_numpy_dtype(self.dtype)(-np.inf),
log_cdf)
def _log_cdf(self, x): return tf.math.log1p(2 / np.pi * tf.atan(self._z(x))) - np.log(2)
def _cdf(self, x): return tf.atan(self._z(x)) / np.pi + 0.5
def _log_cdf(self, x):
return self._extend_support_with_default_value(
x,
lambda x: np.log(2 / np.pi) + tf.math.log(tf.atan(self._z(x))),
default_value=-np.inf)
def _log_cauchy_cdf(z):
return tf.math.log1p(2 / np.pi * tf.atan(z)) - np.log(2)
def _cauchy_cdf(z):
return tf.atan(z) / np.pi + 0.5
