def test_shape_ignore_msg_in_constructor(self):
"""
Ensures that a UserWarning is raised when the 'shape_ref_pos' keyword
argument is passed to the Scobit model constructor. This warns people
against expecting the shape parameters of the Scobit model to suffer
from identification problems. It also alerts users that they are using
a Scobit model when they might have been expecting to instantiate a
different choice model.
"""
# Create a variable for the kwargs being passed to the constructor
kwargs = deepcopy(self.constructor_kwargs)
kwargs["shape_ref_pos"] = 2
# Test to ensure that the shape ignore message is printed when using
# either of these two kwargs
with warnings.catch_warnings(record=True) as context:
# Use this filter to always trigger the UserWarnings
warnings.simplefilter('always', UserWarning)
# Create a Scobit model object with the irrelevant kwargs.
# This should trigger a UserWarning
scobit_obj = scobit.MNSL(*self.constructor_args, **kwargs)
# Check that the warning has been created.
self.assertEqual(len(context), 1)
self.assertIsInstance(context[-1].category, type(UserWarning))
self.assertIn(scobit._shape_ref_msg, str(context[-1].message))
return None
def setUp(self):
# The set up being used is one where there are two choice situations,
# The first having three alternatives, and the second having only two
# alternatives. There is one generic variable. Two alternative
# specific constants and all three shape parameters are used.
# Create the betas to be used during the tests
self.fake_betas = np.array([-0.6])
# Create the fake outside intercepts to be used during the tests
self.fake_intercepts = np.array([1, 0.5])
# Create names for the intercept parameters
self.fake_intercept_names = ["ASC 1", "ASC 2"]
# Record the position of the intercept that is not being estimated
self.fake_intercept_ref_pos = 2
# Create the shape parameters to be used during the tests. Note that
# these are the reparameterized shape parameters, thus they will be
# exponentiated in the fit_mle process and various calculations.
self.fake_shapes = np.array([-1, 0, 1])
# Create names for the intercept parameters
self.fake_shape_names = ["Shape 1", "Shape 2", "Shape 3"]
# Create an array of all model parameters
self.fake_all_params = np.concatenate(
(self.fake_shapes, self.fake_intercepts, self.fake_betas))
# The mapping between rows and alternatives is given below.
self.fake_rows_to_alts = csr_matrix(
np.array([[1, 0, 0], [0, 1, 0], [0, 0, 1], [1, 0, 0], [0, 0, 1]]))
# Create the fake design matrix with columns denoting X
# The intercepts are not included because they are kept outside the
# index in the scobit model.
self.fake_design = np.array([[1], [2], [3], [1.5], [3.5]])
# Create the index array for this set of choice situations
self.fake_index = self.fake_design.dot(self.fake_betas)
# Create the needed dataframe for the Scobit constructor
self.fake_df = pd.DataFrame({
"obs_id": [1, 1, 1, 2, 2],
"alt_id": [1, 2, 3, 1, 3],
"choice": [0, 1, 0, 0, 1],
"x": self.fake_design[:, 0],
"intercept": [1 for i in range(5)]
})
# Record the various column names
self.alt_id_col = "alt_id"
self.obs_id_col = "obs_id"
self.choice_col = "choice"
# Create the index specification and name dictionaryfor the model
self.fake_specification = OrderedDict()
self.fake_names = OrderedDict()
self.fake_specification["x"] = [[1, 2, 3]]
self.fake_names["x"] = ["x (generic coefficient)"]
# Bundle the args and kwargs used to construct the Scobit model.
self.constructor_args = [
self.fake_df, self.alt_id_col, self.obs_id_col, self.choice_col,
self.fake_specification
]
# Create a variable for the kwargs being passed to the constructor
self.constructor_kwargs = {
"intercept_ref_pos": self.fake_intercept_ref_pos,
"names": self.fake_names,
"intercept_names": self.fake_intercept_names,
"shape_names": self.fake_shape_names
}
# Initialize a basic scobit model.
# Create the scobit model object whose coefficients will be estimated.
self.model_obj = scobit.MNSL(*self.constructor_args,
**self.constructor_kwargs)
return None
def make_uneven_and_scobit_models(self):
# The set up being used is one where there are two choice situations,
# The first having three alternatives, and the second having only two
# alternatives. There is one generic variable. Two alternative
# specific constants and all three shape parameters are used.
# Create the betas to be used during the tests
fake_betas = np.array([-0.6])
# Create the fake outside intercepts to be used during the tests
fake_intercepts = np.array([1, 0.5])
# Create names for the intercept parameters
fake_intercept_names = ["ASC 1", "ASC 2"]
# Record the position of the intercept that is not being estimated
fake_intercept_ref_pos = 2
# Create the shape parameters to be used during the tests. Note that
# these are the reparameterized shape parameters, thus they will be
# exponentiated in the fit_mle process and various calculations.
fake_shapes = np.array([-1, 1, 2])
# Create names for the intercept parameters
fake_shape_names = ["Shape 1", "Shape 2", "Shape 3"]
# Create an array of all model parameters
fake_all_params = np.concatenate(
(fake_shapes, fake_intercepts, fake_betas))
# Get the mappping between rows and observations
fake_rows_to_obs = csr_matrix(
np.array([[1, 0], [1, 0], [1, 0], [0, 1], [0, 1]]))
# Create the fake design matrix with columns denoting X
# The intercepts are not included because they are kept outside the
# index in the scobit model.
fake_design = np.array([[1], [2], [3], [1.5], [3.5]])
# Create the index array for this set of choice situations
fake_index = fake_design.dot(fake_betas)
# Create the needed dataframe for the model constructor
fake_df = pd.DataFrame({
"obs_id": [1, 1, 1, 2, 2],
"alt_id": [1, 2, 3, 1, 3],
"choice": [0, 1, 0, 0, 1],
"x": fake_design[:, 0],
"intercept": [1 for i in range(5)]
})
# Record the various column names
alt_id_col = "alt_id"
obs_id_col = "obs_id"
choice_col = "choice"
# Create the index specification and name dictionary for the model
fake_specification = OrderedDict()
fake_names = OrderedDict()
fake_specification["x"] = [[1, 2, 3]]
fake_names["x"] = ["x (generic coefficient)"]
# Bundle args and kwargs used to construct the choice models.
constructor_args = [
fake_df, alt_id_col, obs_id_col, choice_col, fake_specification
]
# Create a variable for the kwargs being passed to the constructor
constructor_kwargs = {
"intercept_ref_pos": fake_intercept_ref_pos,
"names": fake_names,
"intercept_names": fake_intercept_names,
"shape_names": fake_shape_names
}
# Initialize the various choice models
uneven_obj = uneven.MNUL(*constructor_args, **constructor_kwargs)
scobit_obj = scobit.MNSL(*constructor_args, **constructor_kwargs)
for model_obj in [uneven_obj, scobit_obj]:
model_obj.coefs = pd.Series(fake_betas, index=fake_names["x"])
model_obj.intercepts =\
pd.Series(fake_intercepts, index=fake_intercept_names)
model_obj.shapes = pd.Series(fake_shapes, index=fake_shape_names)
model_obj.nests = None
model_obj.params =\
pd.concat([model_obj.shapes,
model_obj.intercepts,
model_obj.coefs],
axis=0, ignore_index=False)
return uneven_obj, scobit_obj