def test_ridge_warning_in_fit_mle(self):
"""
Ensure that a UserWarning is raised when one passes the ridge keyword
argument to the `fit_mle` method of a Nested Logit model object.
"""
# Bundle the arguments used to construct the nested logit model
constructor_args = [
self.fake_df, self.alt_id_col, self.obs_id_col, self.choice_col,
self.fake_specification, self.fake_names
]
# Bundle the kwargs for constructing the nested_logit_model
constructor_kwargs = {"nest_spec": self.fake_nest_spec}
# Create the mnl model object whose coefficients will be estimated.
base_nl = nl.NestedLogit(*constructor_args, **constructor_kwargs)
# Create a variable for the fit_mle function's kwargs.
# The print_res = False arguments are to make sure strings aren't
# printed to the console unnecessarily.
fit_kwargs = {"constrained_pos": [1], "ridge": 0.5, "print_res": False}
# Test to make sure that the ridge warning message is printed when
# using the ridge keyword argument
with warnings.catch_warnings(record=True) as w:
# Use this filter to always trigger the UserWarnings
warnings.simplefilter('always', UserWarning)
base_nl.fit_mle(self.fake_all_params, **fit_kwargs)
self.assertGreaterEqual(len(w), 1)
self.assertIsInstance(w[0].category, type(UserWarning))
self.assertIn(nl._ridge_warning_msg, str(w[0].message))
return None
def test_just_point_kwarg(self):
"""
Ensure that calling `fit_mle` with `just_point = True` returns a
dictionary with a 'x' key and a corresponding value that is an ndarray.
"""
# Bundle the arguments used to construct the nested logit model
constructor_args = [
self.fake_df, self.alt_id_col, self.obs_id_col, self.choice_col,
self.fake_specification
]
# Bundle the kwargs for constructing the nested_logit_model
constructor_kwargs = {
"names": self.fake_names,
"nest_spec": self.fake_nest_spec
}
# Create the mnl model object whose coefficients will be estimated.
base_nl = nl.NestedLogit(*constructor_args, **constructor_kwargs)
# Create a variable for the arguments to the fit_mle function.
fit_args = [self.fake_all_params]
# Alias the function being tested
func = base_nl.fit_mle
# Get the necessary kwargs
kwargs = {"just_point": True}
# Get the function results
func_result = func(*fit_args, **kwargs)
# Perform the desired tests to make sure we get back a dictionary with
# an "x" key in it and a value that is a ndarray.
self.assertIsInstance(func_result, dict)
self.assertIn("x", func_result)
self.assertIsInstance(func_result["x"], np.ndarray)
return None
def test_invalid_init_kwargs_error_in_fit_mle(self):
"""
Ensures that a ValueError is raised when users try to use any other
type of initial value input methods other than the `init_vals`
argument of `fit_mle()`. This prevents people from expecting the use
of outside intercept or shape parameters to work with the Nested Logit
model.
"""
# Bundle the arguments used to construct the nested logit model
constructor_args = [
self.fake_df, self.alt_id_col, self.obs_id_col, self.choice_col,
self.fake_specification
]
# Bundle the kwargs for constructing the nested_logit_model
constructor_kwargs = {
"names": self.fake_names,
"nest_spec": self.fake_nest_spec
}
# Create the mnl model object whose coefficients will be estimated.
base_nl = nl.NestedLogit(*constructor_args, **constructor_kwargs)
# Create a variable for the arguments to the fit_mle function.
# this mimics the arguments passed when trying to use the shape_param
# or outside intercepts kwargs with fit_mle.
fit_args = [None]
# Create variables for the incorrect kwargs.
# The print_res = False arguments are to make sure strings aren't
# printed to the console unnecessarily.
kwarg_map_1 = {"init_shapes": np.array([1, 2]), "print_res": False}
kwarg_map_2 = {"init_intercepts": np.array([1]), "print_res": False}
kwarg_map_3 = {"init_coefs": np.array([1]), "print_res": False}
# Test to ensure that the kwarg ignore message is printed when using
# any of these three incorrect kwargs
for kwargs in [kwarg_map_1, kwarg_map_2, kwarg_map_3]:
self.assertRaises(ValueError, base_nl.fit_mle, *fit_args, **kwargs)
return None
def test_invalid_init_vals_length_in_estimate(self):
"""
Ensure that when _estimate() is called, with an init_values argument
that is of an incorrect length, a ValueError is raised.
"""
# Bundle the arguments used to construct the nested logit model
constructor_args = [
self.fake_df, self.alt_id_col, self.obs_id_col, self.choice_col,
self.fake_specification, self.fake_names
]
# Bundle the kwargs for constructing the nested_logit_model
constructor_kwargs = {"nest_spec": self.fake_nest_spec}
# Create the mnl model object whose coefficients will be estimated.
base_nl = nl.NestedLogit(*constructor_args, **constructor_kwargs)
# Create an estimator object.
zero_vector = np.zeros(self.fake_all_params.shape[0])
estimator_args = [
base_nl,
base_nl.get_mappings_for_fit(), None, zero_vector,
nl.split_param_vec
]
estimator_kwargs = {"constrained_pos": [1]}
nested_estimator = nl.NestedEstimator(*estimator_args,
**estimator_kwargs)
# Alias the function being tested
func = nested_estimator.check_length_of_initial_values
# Test that the desired error is raised
for i in [-1, 1]:
init_values = np.arange(self.fake_all_params.shape[0] + i)
self.assertRaisesRegexp(ValueError,
"values are of the wrong dimension", func,
init_values)
return None
def setUp(self):
# Create the betas to be used during the tests
self.fake_betas = np.array([0.3, -0.6, 0.2])
# Create the fake nest coefficients to be used during the tests
# Note that these are the 'natural' nest coefficients, i.e. the
# inverse of the scale parameters for each nest. They should be less
# than or equal to 1.
self.natural_nest_coefs = np.array([0.995, 0.5])
# Create an array of all model parameters
self.fake_all_params = np.concatenate((self.natural_nest_coefs,
self.fake_betas))
# The set up being used is one where there are two choice situations,
# The first having three alternatives, and the second having only two.
# The nest memberships of these alternatives are given below.
self.fake_rows_to_nests = csr_matrix(np.array([[1, 0],
[1, 0],
[0, 1],
[1, 0],
[0, 1]]))
# Create a sparse matrix that maps the rows of the design matrix to the
# observatins
self.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 ASC_1, ASC_2, X
self.fake_design = np.array([[1, 0, 1],
[0, 1, 2],
[0, 0, 3],
[1, 0, 1.5],
[0, 0, 3.5]])
# Create fake versions of the needed arguments for the MNL 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": range(5),
"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"
# Store the choice array
self.choice_array = self.fake_df[self.choice_col].values
# Create a sparse matrix that maps the chosen rows of the design
# matrix to the observatins
self.fake_chosen_rows_to_obs = csr_matrix(np.array([[0, 0],
[1, 0],
[0, 0],
[0, 0],
[0, 1]]))
# Create the index specification and name dictionaryfor the model
self.fake_specification = OrderedDict()
self.fake_specification["intercept"] = [1, 2]
self.fake_specification["x"] = [[1, 2, 3]]
self.fake_names = OrderedDict()
self.fake_names["intercept"] = ["ASC 1", "ASC 2"]
self.fake_names["x"] = ["x (generic coefficient)"]
# Create the nesting specification
self.fake_nest_spec = OrderedDict()
self.fake_nest_spec["Nest 1"] = [1, 2]
self.fake_nest_spec["Nest 2"] = [3]
# Create a nested logit object
args = [self.fake_df,
self.alt_id_col,
self.obs_id_col,
self.choice_col,
self.fake_specification]
kwargs = {"names": self.fake_names,
"nest_spec": self.fake_nest_spec}
self.model_obj = nested_logit.NestedLogit(*args, **kwargs)
# Store a ridge parameter
self.ridge = 0.5
return None
def make_nested_model(self):
# Create the betas to be used during the tests
fake_betas = np.array([0.3, -0.6, 0.2])
# Create the fake nest coefficients to be used during the tests
# Note that these are the 'natural' nest coefficients, i.e. the
# inverse of the scale parameters for each nest. They should be bigger
# than or equal to 1.
natural_nest_coefs = np.array([1 - 1e-16, 0.5])
# Create an array of all model parameters
fake_all_params = np.concatenate((natural_nest_coefs, fake_betas))
# The set up being used is one where there are two choice situations,
# The first having three alternatives, and the second having only two.
# The nest memberships of these alternatives are given below.
fake_rows_to_nests = csr_matrix(
np.array([[1, 0], [1, 0], [0, 1], [1, 0], [0, 1]]))
# Create a sparse matrix that maps the rows of the design matrix to the
# observatins
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 ASC_1, ASC_2, X
fake_design = np.array([[1, 0, 1], [0, 1, 2], [0, 0, 3], [1, 0, 1.5],
[0, 0, 3.5]])
# Create fake versions of the needed arguments for the MNL 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": range(5),
"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"
# Store the choice array
choice_array = fake_df[choice_col].values
# Create a sparse matrix that maps the chosen rows of the design
# matrix to the observatins
fake_chosen_rows_to_obs = csr_matrix(
np.array([[0, 0], [1, 0], [0, 0], [0, 0], [0, 1]]))
# Create the index specification and name dictionaryfor the model
fake_specification = OrderedDict()
fake_specification["intercept"] = [1, 2]
fake_specification["x"] = [[1, 2, 3]]
fake_names = OrderedDict()
fake_names["intercept"] = ["ASC 1", "ASC 2"]
fake_names["x"] = ["x (generic coefficient)"]
# Create the nesting specification
fake_nest_spec = OrderedDict()
fake_nest_spec["Nest 1"] = [1, 2]
fake_nest_spec["Nest 2"] = [3]
# Create a nested logit object
args = [
fake_df, alt_id_col, obs_id_col, choice_col, fake_specification
]
kwargs = {"names": fake_names, "nest_spec": fake_nest_spec}
model_obj = nested_logit.NestedLogit(*args, **kwargs)
model_obj.coefs = pd.Series(fake_betas, index=model_obj.ind_var_names)
model_obj.intercepts = None
model_obj.shapes = None
def logit(x):
return np.log(x / (1 - x))
model_obj.nests =\
pd.Series(logit(natural_nest_coefs), index=fake_nest_spec.keys())
model_obj.params =\
pd.concat([model_obj.nests, model_obj.coefs],
axis=0, ignore_index=False)
return model_obj
