def test__extract_constant(self):
distribution = GaussianUnivariate()
distribution._params = {'loc': 1, 'scale': 0}
constant = distribution._extract_constant()
assert 1 == constant
def test__fit_constant(self):
distribution = GaussianUnivariate()
distribution._fit_constant(np.array([1, 1, 1, 1]))
assert distribution._params == {
'loc': 1,
'scale': 0
}
def test_fit_empty_data(self):
"""On fit, if column is empty an error is raised."""
# Setup
copula = GaussianUnivariate()
column = pd.Series([])
# Run
with self.assertRaises(ValueError):
copula.fit(column)
def test_valid_serialization_unfit_model(self):
"""For a unfitted model to_dict and from_dict are opposites."""
# Setup
instance = GaussianUnivariate()
# Run
result = GaussianUnivariate.from_dict(instance.to_dict())
# Check
assert instance.to_dict() == result.to_dict()
def test__fit(self):
distribution = GaussianUnivariate()
data = norm.rvs(size=1000, loc=1, scale=1)
distribution._fit(data)
assert distribution._params == {
'loc': np.mean(data),
'scale': np.std(data),
}
def test_test_fit_equal_values(self):
"""On fit, even if column has equal values, std is never 0."""
# Setup
copula = GaussianUnivariate()
column = [1, 1, 1, 1, 1, 1]
# Run
copula.fit(column)
# Check
assert copula.mean == 1
assert copula.std == 0.001
def test_to_dict(self):
"""To_dict returns the defining parameters of a distribution in a dict."""
# Setup
copula = GaussianUnivariate()
column = [0, 1, 2, 3, 4, 5]
copula.fit(column)
expected_result = {'mean': 2.5, 'std': 1.707825127659933}
# Run
result = copula.to_dict()
# Check
assert result == expected_result
def test_sample(self):
"""After fitting, GaussianUnivariate is able to sample new data."""
# Setup
copula = GaussianUnivariate()
column = [-1, 0, 1]
copula.fit(column)
# Run
result = copula.sample(1000000)
# Check
assert len(result) == 1000000
assert abs(np.mean(result) - copula.mean) < 10E-3
assert abs(np.std(result) - copula.std) < 10E-3
def test_test_fit_equal_values(self):
"""If it's fit with constant data, contant_value is set."""
# Setup
instance = GaussianUnivariate()
column = np.array([5, 5, 5, 5, 5, 5])
# Run
instance.fit(column)
# Check
assert instance.mean == 0
assert instance.std == 1
assert instance.constant_value == 5
def test_get_probability_density(self):
"""Probability_density returns the normal probability distribution for the given values."""
# Setup
copula = GaussianUnivariate()
column = np.array([-1, 0, 1])
copula.fit(column)
expected_result = 0.48860251190292
# Run
result = copula.probability_density(0)
# Check
assert result == expected_result
def test_sample_random_state(self):
"""When random state is set, samples are the same."""
# Setup
instance = GaussianUnivariate(random_seed=0)
X = np.array([1, 2, 3, 4, 5])
instance.fit(X)
expected_result = np.array([5.494746752403546, 3.565907751154284, 4.384144531132039])
# Run
result = instance.sample(3)
# Check
assert (result == expected_result).all()
def test_percent_point(self):
"""Percent_point returns the original point from the cumulative probability value."""
# Setup
copula = GaussianUnivariate()
column = np.array([-1, 0, 1])
copula.fit(column)
x = 0.5
expected_result = 0
# Run
result = copula.percent_point(x)
# Check
assert (result == expected_result).all()
def test_cumulative_distribution(self):
"""Cumulative_density returns the cumulative distribution value for a point."""
# Setup
copula = GaussianUnivariate()
column = np.array([-1, 0, 1])
copula.fit(column)
x = pd.Series([0])
expected_result = [0.5]
# Run
result = copula.cumulative_distribution(x)
# Check
assert (result == expected_result).all()
def test___str__(self):
"""str returns details about the model."""
# Setup
copula = GaussianUnivariate()
expected_result = '\n'.join([
'Distribution Type: Gaussian', 'Variable name: None', 'Mean: 0',
'Standard deviation: 1'
])
# Run
result = copula.__str__()
# Check
assert result == expected_result
def test_fit(self):
"""On fit, stats from fit data are set in the model."""
# Setup
copula = GaussianUnivariate()
column = pd.Series([0, 1, 2, 3, 4, 5], name='column')
# Run
copula.fit(column)
# Check
assert copula.mean == 2.5
assert copula.std == 1.707825127659933
assert copula.name == 'column'
assert copula.fitted is True
def fit(self, X, distrib_map=None):
"""Compute the distribution for each variable and then its covariance matrix.
Args:
X: `numpy.ndarray` or `pandas.DataFrame`. Data to model.
distrib_map: `dict` mapping of distributions for the columns in X.
Returns:
None
"""
LOGGER.debug('Fitting Gaussian Copula')
column_names = self.get_column_names(X)
# create distributions based on user input
if distrib_map:
for key in distrib_map:
# this isn't fully working yet
self.distribs[key] = distrib_map[key](X[key])
else:
for column_name in column_names:
self.distribs[column_name] = GaussianUnivariate()
column = self.get_column(X, column_name)
self.distribs[column_name].fit(column)
self.covariance = self._get_covariance(X)
def test_to_dict(self):
"""To_dict returns the defining parameters of a distribution in a dict."""
# Setup
copula = GaussianUnivariate()
column = np.array([0, 1, 2, 3, 4, 5])
copula.fit(column)
expected_result = {
'type': 'copulas.univariate.gaussian.GaussianUnivariate',
'mean': 2.5,
'std': 1.707825127659933,
'fitted': True,
}
# Run
result = copula.to_dict()
# Check
assert result == expected_result
def from_dict(cls, copula_dict):
"""Set attributes with provided values."""
instance = cls()
instance.distribs = {}
for name, parameters in copula_dict['distribs'].items():
instance.distribs[name] = GaussianUnivariate.from_dict(parameters)
instance.covariance = np.array(copula_dict['covariance'])
return instance
def test_sample(self, random_mock):
"""After fitting, GaussianUnivariate is able to sample new data."""
# Setup
instance = GaussianUnivariate()
column = np.array([-1, 0, 1])
instance.fit(column)
expected_result = np.array([1, 2, 3, 4, 5])
random_mock.return_value = expected_result
# Run
result = instance.sample(5)
# Check
compare_nested_iterables(result, expected_result)
assert instance.mean == 0.0
assert instance.std == 0.816496580927726
random_mock.assert_called_once_with(0.0, 0.816496580927726, 5)
def test___init__(self):
"""On init, default values are set on instance."""
# Setup / Run
copula = GaussianUnivariate()
# Check
assert not copula.name
assert copula.mean == 0
assert copula.std == 1
def test_from_dict(self):
"""From_dict sets the values of a dictionary as attributes of the instance."""
# Setup
parameters = {'mean': 2.5, 'std': 1.707825127659933}
# Run
copula = GaussianUnivariate.from_dict(parameters)
# Check
assert copula.mean == 2.5
assert copula.std == 1.707825127659933
copula.sample(10)
def test_percent_point_reverse_cumulative_distribution(self):
"""Combined cumulative_distribution and percent_point is the identity function."""
# Setup
copula = GaussianUnivariate()
column = np.array([-1, 0, 1])
copula.fit(column)
initial_value = np.array([0, 0.1, 0.2, 0.3, 0.4, 0.5])
# Run
result_a = copula.percent_point(copula.cumulative_distribution(initial_value))
result_b = copula.cumulative_distribution(copula.percent_point(initial_value))
# Check
assert (initial_value - result_a < 10E-5).all()
assert (initial_value - result_b < 10E-5).all()
def test_percent_point_reverse_cumulative_distribution(self):
"""Combined cumulative_distribution and percent_point is the identity function."""
# Setup
copula = GaussianUnivariate()
column = [-1, 0, 1]
copula.fit(column)
initial_value = pd.Series([0])
# Run
result_a = copula.percent_point(
copula.cumulative_distribution(initial_value))
result_b = copula.cumulative_distribution(
copula.percent_point(initial_value))
# Check
assert (initial_value == result_a).all()
assert (initial_value == result_b).all()
def test_sample_constant(self):
"""samples can be generated for constant distribution."""
# Setup
instance = GaussianUnivariate()
instance.constant_value = 3
instance._replace_constant_methods()
instance.fitted = True
expected_result = np.array([3, 3, 3, 3, 3])
# Run
result = instance.sample(5)
# Check
compare_nested_iterables(result, expected_result)
def test_cumulative_distribution_constant(self):
"""cumulative_distribution can be computed for constant distribution."""
# Setup
instance = GaussianUnivariate()
instance.constant_value = 3
instance._replace_constant_methods()
instance.fitted = True
X = np.array([1, 2, 3, 4, 5])
expected_result = np.array([0, 0, 1, 1, 1])
# Run
result = instance.cumulative_distribution(X)
# Check
compare_nested_iterables(result, expected_result)
def test__is_constant_false(self):
distribution = GaussianUnivariate()
distribution.fit(np.array([1, 2, 3, 4]))
assert not distribution._is_constant()
def test__is_constant_true(self):
distribution = GaussianUnivariate()
distribution.fit(np.array([1, 1, 1, 1]))
assert distribution._is_constant()
