def covariance_stuff(self, cost_function):
ls_zero_step = scipy_ls(cost_function, self.fit_coefficients,
args=(self.data_frame.temp, self.data_frame.experiment))
temporary_parameters = ls_zero_step.x
jacobian = ls_zero_step.jac
covariance_matrix = np.linalg.inv(jacobian.T.dot(jacobian))
temporary_fit = cost_function(temporary_parameters, self.data_frame.temp, self.data_frame.experiment)
temporary_fit *= temporary_fit
# todo read https://stackoverflow.com/questions/28702631/scipy-curve-fit-returns-negative-variance
sigma_squared = sum(temporary_fit) / (len(self.data_frame.temp) - len(self.fit_coefficients) / 2)
covariance_matrix *= sigma_squared
diagonal = np.sqrt(np.diagonal(covariance_matrix))
self.confidence_params = np.sqrt(diagonal)
self.confidence_params *= np.abs(self.level_t())
print('Confidence something: ', self.confidence_params)
need_new = False
for a, b in zip(self.fit_coefficients, self.confidence_params):
if a < b:
need_new = True
if need_new:
self.fit_coefficients = np.append(self.fit_coefficients, 0.1)
self.fit_coefficients = np.append(self.fit_coefficients, 10)
def j_mode_approximation(self):
t_ref = self.data_frame.reference_temperature
t_ref_vector = t_ref * np.ones(len(self.enthalpy_temperature))
c_0 = self.data_frame.reference_enthalpy_value
c_1 = self.data_frame.reference_heat_capacity_value
updated_enthalpy = self.enthalpy_data - c_0 * np.ones(
len(self.enthalpy_temperature)) - c_1 * (
self.enthalpy_temperature - t_ref_vector)
updated_heat_capacity = self.data_frame.cp_e - c_1 * np.ones(
len(self.heat_capacity_temperature))
initial_fit = scipy_ls(self.joint_cost_function,
self.params,
args=(self.enthalpy_temperature,
updated_enthalpy, self.data_frame.cp_t,
updated_heat_capacity))
self.fit_coefficients = self.stationary_coefficients(
initial_fit.x.tolist(), t_ref, c_0, c_1)
self.fit_enthalpy = self.delta_enthalpy(self.fit_coefficients,
self.data_frame.dh_t)
self.fit_heat_capacity = self.heat_capacity(self.fit_coefficients,
self.data_frame.cp_t)
def approx(self):
if self.mode == 'j':
res_lsq = scipy_ls(self.joint_cost_function, self.params, # bounds=self.bounds,
args=(self.data_frame.dh_t, self.data_frame.dh_e,
self.data_frame.cp_t, self.data_frame.cp_e))
elif self.mode == 'h':
res_lsq = scipy_ls(self.delta_enthalpy_cost, self.params, # bounds=self.bounds,
args=(self.data_frame.dh_t, self.data_frame.dh_e))
elif self.mode == 'c':
res_lsq = scipy_ls(self.heat_capacity_cost, self.params, # bounds=self.bounds,
args=(self.data_frame.cp_t, self.data_frame.cp_e))
else:
print('Wrong calculation type!')
raise ValueError('Type can only be h c j.\n')
# print(res_lsq.x.tolist())
self.params = res_lsq.x.tolist()
def c_mode_approximation(self):
self.params = np.ones(self.max_power - self.min_power + 1)
initial_fit = scipy_ls(self.heat_capacity_cost,
self.params,
args=(self.data_frame.cp_t,
self.data_frame.cp_e))
self.fit_coefficients = initial_fit.x.tolist()
# self.fit_coefficients = self.stationary_coefficients(initial_fit.x.tolist(), t_ref, c_0, c_1)
self.fit_enthalpy = self.delta_enthalpy(self.fit_coefficients,
self.data_frame.dh_t)
self.fit_heat_capacity = self.heat_capacity(self.fit_coefficients,
self.data_frame.cp_t)
def approx(self):
for i in range(0, self.power - 1):
self.fit_coefficients.append(0.01)
self.fit_coefficients.append(1.00)
self.bounds[0].append(0.0)
self.bounds[0].append(0.0)
self.bounds[1].append(500.0)
self.bounds[1].append(1.0e5)
if self.mode == 'j':
res_lsq = scipy_ls(self.joint_cost_function, self.fit_coefficients, bounds=self.bounds,
args=(self.data_frame.dh_t, self.data_frame.dh_e,
self.data_frame.cp_t, self.data_frame.cp_e))
elif self.mode == 'h':
res_lsq = scipy_ls(self.delta_enthalpy_cost, self.fit_coefficients, bounds=self.bounds,
args=(self.data_frame.dh_t, self.data_frame.dh_e))
elif self.mode == 'c':
res_lsq = scipy_ls(self.heat_capacity_cost, self.fit_coefficients, bounds=self.bounds,
args=(self.data_frame.cp_t, self.data_frame.cp_e))
else:
print('Wrong calculation type!')
raise ValueError('Type can only be h c j.\n')
self.fit_coefficients = res_lsq.x.tolist()
def c_mode_constrained_approximation(self):
# todo change to appr, add another named unconstrained
t_ref = self.data_frame.reference_temperature
c_0 = self.data_frame.reference_enthalpy_value
c_1 = self.data_frame.reference_heat_capacity_value
updated_cp = self.data_frame.cp_e - c_1 * np.ones(len(self.heat_capacity_temperature))
initial_fit = scipy_ls(self.heat_capacity_constrained_cost, self.params,
args=(self.data_frame.cp_t, updated_cp))
self.fit_coefficients = self.stationary_coefficients(initial_fit.x.tolist(), t_ref, c_0, c_1)
self.fit_enthalpy = self.delta_enthalpy(self.fit_coefficients, self.data_frame.dh_t)
self.fit_heat_capacity = self.heat_capacity(self.fit_coefficients, self.data_frame.cp_t)
def h_mode_approximation(self):
t_ref = self.data_frame.reference_temperature
t_ref_vector = t_ref * np.ones(len(self.enthalpy_temperature))
c_0 = self.data_frame.reference_enthalpy_value
c_1 = self.data_frame.reference_heat_capacity_value
updated_experiment = self.data_frame.dh_e - c_0 * np.ones(
len(self.enthalpy_temperature)) - c_1 * (
self.enthalpy_temperature - t_ref_vector)
initial_fit = scipy_ls(self.delta_enthalpy_constrained_cost,
self.params,
args=(self.data_frame.dh_t, updated_experiment))
print(self.name, initial_fit.x.tolist())
self.fit_coefficients = self.stationary_coefficients(
initial_fit.x.tolist(), t_ref, c_0, c_1)
self.fit_enthalpy = self.delta_enthalpy(self.fit_coefficients,
self.data_frame.dh_t)
self.fit_heat_capacity = self.heat_capacity(self.fit_coefficients,
self.data_frame.cp_t)