Python special.log_ndtr方法代码示例

# 需要导入模块: from scipy import special [as 别名]
# 或者: from scipy.special import log_ndtr [as 别名]
def _test_grid_log(self, dtype, grid_spec, error_spec):
    with self.test_session():
      grid = _make_grid(dtype, grid_spec)
      actual = sm.log_ndtr(grid).eval()

      # Basic tests.
      self.assertTrue(np.isfinite(actual).all())
      # On the grid, -inf < log_cdf(x) < 0.  In this case, we should be able
      # to use a huge grid because we have used tricks to escape numerical
      # difficulties.
      self.assertTrue((actual < 0).all())
      _check_strictly_increasing(actual)

      # Versus scipy.
      expected = special.log_ndtr(grid)
      # Scipy prematurely goes to zero at some places that we don't.  So don't
      # include these in the comparison.
      self.assertAllClose(expected.astype(np.float64)[expected < 0],
                          actual.astype(np.float64)[expected < 0],
                          rtol=error_spec.rtol, atol=error_spec.atol) 

# 需要导入模块: from scipy import special [as 别名]
# 或者: from scipy.special import log_ndtr [as 别名]
def _log_erfc(x):
  """Compute log(erfc(x)) with high accuracy for large x."""
  try:
    return math.log(2) + special.log_ndtr(-x * 2**.5)
  except NameError:
    # If log_ndtr is not available, approximate as follows:
    r = special.erfc(x)
    if r == 0.0:
      # Using the Laurent series at infinity for the tail of the erfc function:
      #     erfc(x) ~ exp(-x^2-.5/x^2+.625/x^4)/(x*pi^.5)
      # To verify in Mathematica:
      #     Series[Log[Erfc[x]] + Log[x] + Log[Pi]/2 + x^2, {x, Infinity, 6}]
      return (-math.log(math.pi) / 2 - math.log(x) - x**2 - .5 * x**-2 +
              .625 * x**-4 - 37. / 24. * x**-6 + 353. / 64. * x**-8)
    else:
      return math.log(r) 

# 需要导入模块: from scipy import special [as 别名]
# 或者: from scipy.special import log_ndtr [as 别名]
def _norm_logcdf(x):
    return sc.log_ndtr(x) 

# 需要导入模块: from scipy import special [as 别名]
# 或者: from scipy.special import log_ndtr [as 别名]
def _log_ndtr_cpu(x, dtype):
    from scipy import special
    return special.log_ndtr(x).astype(dtype) 

# 需要导入模块: from scipy import special [as 别名]
# 或者: from scipy.special import log_ndtr [as 别名]
def label(self):
        return 'log_ndtr' 

# 需要导入模块: from scipy import special [as 别名]
# 或者: from scipy.special import log_ndtr [as 别名]
def forward_cpu(self, x):
        global _log_ndtr_cpu
        if _log_ndtr_cpu is None:
            try:
                from scipy import special
                _log_ndtr_cpu = special.log_ndtr
            except ImportError:
                raise ImportError('SciPy is not available. Forward computation'
                                  ' of log_ndtr can not be done.')

        self.retain_inputs((0,))
        return utils.force_array(_log_ndtr_cpu(x[0]), dtype=x[0].dtype), 

# 需要导入模块: from scipy import special [as 别名]
# 或者: from scipy.special import log_ndtr [as 别名]
def log_ndtr(x):
    """Logarithm of cumulative distribution function of normal distribution.

    .. note::
       Forward computation in CPU can not be done if
       `SciPy <https://www.scipy.org/>`_ is not available.

    Args:
        x (:class:`~chainer.Variable` or :ref:`ndarray`): Input variable.

    Returns:
        ~chainer.Variable: Output variable.
    """
    return LogNdtr().apply((x,))[0] 

# 需要导入模块: from scipy import special [as 别名]
# 或者: from scipy.special import log_ndtr [as 别名]
def test_log_ndtr(self):
        assert_mpmath_equal(sc.log_ndtr,
                            exception_to_nan(lambda z: mpmath.log(mpmath.ncdf(z))),
                            [Arg()], n=600, dps=300) 

# 需要导入模块: from scipy import special [as 别名]
# 或者: from scipy.special import log_ndtr [as 别名]
def test_log_ndtr_complex(self):
        assert_mpmath_equal(sc.log_ndtr,
                            exception_to_nan(lambda z: mpmath.log(mpmath.erfc(-z/np.sqrt(2.))/2.)),
                            [ComplexArg(a=complex(-10000, -100),
                                        b=complex(10000, 100))], n=200, dps=300) 

# 需要导入模块: from scipy import special [as 别名]
# 或者: from scipy.special import log_ndtr [as 别名]
def _test_grad_finite(self, dtype):
    with self.test_session():
      x = tf.Variable([-100., 0., 100.], dtype=dtype)
      output = (sm.log_ndtr(x) if self._use_log else sm.ndtr(x))
      grad_output = tf.gradients(output, x)
      tf.global_variables_initializer().run()
      self.assert_all_true(np.isfinite(output.eval()))
      self.assert_all_true(np.isfinite(grad_output[0].eval())) 

# 需要导入模块: from scipy import special [as 别名]
# 或者: from scipy.special import log_ndtr [as 别名]
def _log_erfc(x):
    """Compute log(erfc(x)) with high accuracy for large x."""
    return math.log(2) + special.log_ndtr(-x * 2 ** 0.5) 

# 需要导入模块: from scipy import special [as 别名]
# 或者: from scipy.special import log_ndtr [as 别名]
def _test_grad_accuracy(self, dtype, grid_spec, error_spec):
    raw_grid = _make_grid(dtype, grid_spec)
    grid = tf.convert_to_tensor(raw_grid)
    with self.test_session():
      fn = sm.log_ndtr if self._use_log else sm.ndtr

      # If there are N points in the grid,
      # grad_eval.shape = (N, N), with grad_eval[i, j] the partial derivative of
      # the ith output point w.r.t. the jth grid point.  We only expect the
      # diagonal to be nonzero.
      # TODO(b/31131137): Replace tf.test.compute_gradient with our own custom
      # gradient evaluation to ensure we correctly handle small function delta.
      grad_eval, _ = tf.test.compute_gradient(
          grid, grid_spec.shape, fn(grid), grid_spec.shape)
      grad_eval = np.diag(grad_eval)

      # Check for NaN separately in order to get informative failures.
      self.assert_all_false(np.isnan(grad_eval))
      self.assert_all_true(grad_eval > 0.)
      self.assert_all_true(np.isfinite(grad_eval))

      # Do the same checks but explicitly compute the gradient.
      # (We did this because we're not sure if we trust
      # tf.test.compute_gradient.)
      grad_eval = tf.gradients(fn(grid), grid)[0].eval()
      self.assert_all_false(np.isnan(grad_eval))
      if self._use_log:
        g = np.reshape(grad_eval, [-1])
        half = np.ceil(len(g)/2)
        self.assert_all_true(g[:half] > 0.)
        self.assert_all_true(g[half:] >= 0.)
      else:
        # The ndtr gradient will only be non-zero in the range [-14, 14] for
        # float32 and [-38, 38] for float64.
        self.assert_all_true(grad_eval >= 0.)
      self.assert_all_true(np.isfinite(grad_eval))

      # Versus scipy.
      expected = stats.norm.pdf(raw_grid)
      if self._use_log:
        expected /= special.ndtr(raw_grid)
        expected[np.isnan(expected)] = 0.
      # Scipy prematurely goes to zero at some places that we don't.  So don't
      # include these in the comparison.
      self.assertAllClose(expected.astype(np.float64)[expected < 0],
                          grad_eval.astype(np.float64)[expected < 0],
                          rtol=error_spec.rtol, atol=error_spec.atol)