class LeastSquaresFit:
def __init__(self, model, config, param_dict, logger):
self.config = config
self.model = model
self.res = Residuum(model)
self.callable = FitWrapper(self.res.getResiduals())
self.symbols = model.get_Symbols()
self.param_dict = param_dict
self.logger = logger
def fit(self):
self.logger.info("Symbols: %s" % self.symbols)
self.logger.info('Method %s' % self.config['method'])
bounds = tuple(map(list, zip(*self.config['limits']))) # [(min,max),(min1,max1)..] -> ([min,min1,..], [max,max1..])
self.result = least_squares(self.callable, self.config['initial_values'], args=(self.param_dict['spectra_range'],), bounds=bounds)
self.param_dict['variables'] = dict()
for idx, symbol in enumerate(self.symbols):
self.param_dict['variables'][symbol] = dict()
self.param_dict['variables'][symbol]['stderr'] = None
self.param_dict['variables'][symbol]['value'] = self.result.x[idx]
self._calc_fitted_spectra()
self._calc_residuals()
return Result(self.result, self.fitted_spectra, self.residuals, self.param_dict['wave_range']), self.param_dict
def _calc_fitted_spectra(self):
f = self.model.get_compiled()
self.fitted_spectra = f(*self.result.x)
def _calc_residuals(self):
#TODO
self.residuals = self.param_dict['spectra_range'] - self.fitted_spectra
def _iterate(simulation, variables, expected, guess, bounds, biased_parameter, model, input_values, logger, len_parameter=3):
biased_parameters = np.zeros((len(input_values), len_parameter))
for i, input_ in enumerate(input_values):
model.setVariable(biased_parameter, input_)
logger.debug(">>>> Input for %s: %s" % (biased_parameter, input_))
logger.debug("Fit variables %s: " % model.get_Symbols())
res = Residuum(model, 'ratio')
residuals = FitWrapper(res.getResiduals())
Fit = FitModel()
resultls = Fit._least_squares(residuals, guess, simulation, bounds)
biased_parameters[i] = np.array(resultls.x) - np.array(expected)
return biased_parameters
def test_main():
from get_ssa import get_ssa
zenith = 53.1836240528
AMass = 1.66450160404
rel_h = 0.665
pressure = 950
AM = 5
ssa = get_ssa(rel_h, AM)
x = np.linspace(200, 800, 100) # config
variables = ['alpha', 'beta', 'g_dsa', 'g_dsr'] # config
expected_values = [2.5, 0.06, 0.6, 0.5]
print('Expected: %s' % expected_values)
guess = [1.0, 0.01, 0.5, 0.8] # config
bounds = [(-0.2, 4), (0., 3), (0., 2.), (0., 2.)] # config
# Theano
irr_symbol = IrradianceModel_sym(x, zenith, AMass, pressure, ssa, variables)
getIrrRatio = irr_symbol.getcompiledModel('ratio')
y_theano = getIrrRatio(*expected_values)
res = Residuum(irr_symbol, 'ratio')
residuum = FitWrapper(res.getResiduum())
residuals = FitWrapper(res.getResiduals())
derivative = FitWrapper(res.getDerivative())
Fit = FitModel()
result = Fit._minimize(residuum, guess, y_theano, bounds, jacobian=derivative)
print("Got %s" % result.x)
resultls = Fit._least_squares(residuals, guess, y_theano, bounds)
print("Got %s" % resultls.x)
# Python
IrradianceObject = IrradianceModel_python(AMass, rel_h, ssa, zenith, pressure)
y_python = IrradianceObject.irradiance_ratio(x, 2.5, 0.06,0.0, 0.6, 0.5)
gmod = Model(IrradianceObject.irradiance_ratio, independent_vars=['x'], param_names=variables)
gmod.set_param_hint('alpha', value=guess[0], min=bounds[0][0], max=bounds[0][1])
gmod.set_param_hint('beta', value=guess[1], min=bounds[1][0], max=bounds[1][1])
gmod.set_param_hint('g_dsa', value=guess[2], min=bounds[2][0], max=bounds[2][1])
gmod.set_param_hint('g_dsr', value=guess[3], min=bounds[3][0], max=bounds[3][1])
result_lmfit = gmod.fit(y_python, x=x)
print(result_lmfit.fit_report())
plt.plot(x, y_theano)
x_new = np.linspace(300, 900,150)
irr_symbol.set_wavelengthAOI(x_new)
getIrrRatio = irr_symbol.getcompiledModel('ratio')
y_new = getIrrRatio(*expected_values)
plt.plot(x_new, y_new, '+', label='different wavelengths')
plt.legend()
plt.show()
