def _rescaled_cvar(self, x):
return C_CALL.empirical_cvar_py_interface(
np.int32(x), np.int32(self.n), np.int32(self.gen.min),
np.int32(self.gen.max), ffi.cast("int *",
self.nd_vals.ctypes.data),
ffi.cast("double *", self.gen.cdf_vals.ctypes.data),
ffi.cast("double *", self.gen.expectation_vals.ctypes.data))
def _rescaled_cvar(self, x):
cvar = C_CALL.semiparametric_cvar_py_interface(
np.int32(x), np.float64(self.u), np.float64(self.p),
np.float64(self.sigma), np.float64(self.xi), np.int32(self.n),
np.int32(self.gen.min), np.int32(self.gen.max),
ffi.cast("int *", self.nd_vals.ctypes.data),
ffi.cast("double *", self.gen.cdf_vals.ctypes.data),
ffi.cast("double *", self.gen.expectation_vals.ctypes.data))
if cvar == -1:
cvar = np.Inf
return cvar
def cdf(self, m):
"""calculate margin CDF values
**Parameters**:
`m` (`float`): point to evaluate on
"""
return C_CALL.semiparametric_power_margin_cdf_py_interface(
np.int32(m), np.float64(self.u), np.float64(self.p),
np.float64(self.sigma), np.float64(self.xi), np.int32(self.n),
np.int32(self.gen.min), np.int32(self.gen.max),
ffi.cast("int *", self.nd_vals.ctypes.data),
ffi.cast("double *", self.gen.cdf_vals.ctypes.data))
def cdf(self, x):
"""calculate margin CDF values
**Parameters**:
`x` (`numpy.ndarray`): point to evaluate on
"""
if x >= self.max:
return 1.0
elif x < self.min:
return 0.0
else:
return C_CALL.empirical_power_margin_cdf_py_interface(
np.int32(x), np.int32(self.n), np.int32(self.gen.min),
np.int32(self.gen.max),
ffi.cast("int *", self.nd_vals.ctypes.data),
ffi.cast("double *", self.gen.cdf_vals.ctypes.data))
def cvar_trace(self, x):
"""calculate cvar values for each posterior parameter sample
**Parameters**:
`x` (`float`): absolute value of power margin shortfall's lower bound
"""
if x < 0:
raise Exception("x must be a non-negative number.")
output = np.ascontiguousarray(np.empty((self.n_posterior, )),
dtype=np.float64)
C_CALL.bayesian_semiparametric_cvar_trace_py_interface(
np.int32(x), np.float64(self.u), np.float64(self.p),
np.int32(self.n_posterior),
ffi.cast("double *", self.posterior_sigma.ctypes.data),
ffi.cast("double *", self.posterior_xi.ctypes.data),
np.int32(self.n), np.int32(self.gen.min), np.int32(self.gen.max),
ffi.cast("int *", self.nd_vals.ctypes.data),
ffi.cast("double *", self.gen.cdf_vals.ctypes.data),
ffi.cast("double *", self.gen.expectation_vals.ctypes.data),
ffi.cast("double *", output.ctypes.data))
if output[0] == -1:
raise Exception(
"cvar is unbounded; some posterior shape samples are larger than 1"
)
else:
return output
def cdf_trace(self, x):
"""calculate margin CDF values for each posterior parameter sample
**Parameters**:
`x` (`float`): point to evaluate on
"""
output = np.ascontiguousarray(np.empty((self.n_posterior, )),
dtype=np.float64)
C_CALL.bayesian_semiparametric_power_margin_cdf_trace_py_interface(
np.int32(m), np.float64(self.u), np.float64(self.p),
np.int32(self.n_posterior),
ffi.cast("double *", self.posterior_sigma.ctypes.data),
ffi.cast("double *", self.posterior_xi.ctypes.data),
np.int32(self.n), np.int32(self.gen.min), np.int32(self.gen.max),
ffi.cast("int *", self.nd_vals.ctypes.data),
ffi.cast("double *", self.gen.cdf_vals.ctypes.data),
ffi.cast("double *", output.ctypes.data))
return output
def __init__(self, gen, nd_data, u, sigma, xi):
super().__init__(gen, nd_data)
self.gen = gen
self.n = len(self.nd_vals)
self.u = u
self.sigma = sigma
#if np.abs(xi) < 1e-8: #to avoid numerical instability
# xi = 0
self.xi = xi
self.p = C_CALL.empirical_net_demand_cdf_py_interface(
np.float64(self.u), np.int32(self.n),
ffi.cast("int *", self.nd_vals.ctypes.data))