def run(self, data, *args, **kwargs):
self._last_coefficients = None
result = bhhh_mnl_estimation.run(self, data, *args, **kwargs)
# run a diagnose procedure
if self._last_coefficients is not None:
utilities_class = linear_utilities_diagnose()
utilities_class.run(data, self._last_coefficients, resources=self.resources)
return result
def run(self, data, upc_sequence, resources):
"""
'data' is of shape (nobservations, nchoices, nvariables).
WESML procedure (Weighted endogenous sampling maximum likelihood), Manski, Lerman 1977
data are weighted by correction weights (observation_share/sampled_share) in order to take into account
undersampled or oversampled observations.
Set the variable name for the correction weights into estimate_config (in choice_model) as
"wesml_sampling_correction_variable" entry.
"""
model = resources.get("_model_")
self.wesml_weights = model.get_attribute_for_submodel(resources.get("wesml_sampling_correction_variable"), resources)
return bhhh_mnl_estimation.run(self, data, upc_sequence, resources)
def run(self, data, upc_sequence, resources):
"""
'data' is of shape (nobservations, nchoices, nvariables).
WESML procedure (Weighted endogenous sampling maximum likelihood), Manski, Lerman 1977
data are weighted by correction weights (observation_share/sampled_share) in order to take into account
undersampled or oversampled observations.
Set the variable name for the correction weights into estimate_config (in choice_model) as
"wesml_sampling_correction_variable" entry.
"""
model = resources.get("_model_")
self.wesml_weights = model.get_attribute_for_submodel(
resources.get("wesml_sampling_correction_variable"), resources)
return bhhh_mnl_estimation.run(self, data, upc_sequence, resources)
def run(self, data, upc_sequence, resources):
"""
'data' is of shape (nobservations, nchoices, nvariables).
WESML procedure (Weighted endogenous sampling maximum likelihood), Manski, Lerman 1977
data are weighted by correction weights (observation_share/sampled_share) in order to take into account
undersampled or oversampled observations.
Set the variable name for the correction weights into estimate_config (in choice_model) as
"wesml_sampling_correction_variable" entry.
"""
model = resources.get("_model_")
self.wesml_weights = model.choice_set.compute_variables([resources.get("wesml_sampling_correction_variable")],
dataset_pool=model.dataset_pool)
self.wesml_weights = self.wesml_weights[resources.get("index")]
weighted_data = data*reshape(self.wesml_weights, (data.shape[0], data.shape[1], 1))
return bhhh_mnl_estimation.run(self, weighted_data, upc_sequence, resources)
