add more tests

multi_imbalance/resampling/gmm_sampler.py

@@ -279,7 +279,7 @@ def _construct_neighborhood(self, X: np.ndarray, y: np.ndarray) -> None:
  def _check_sample_neighborhood(self, sample_class: int, neigh_counts: Counter[int]) -> float:
  neighborhood = 0.0
  for neigh_class, count in neigh_counts.items():
- class_sizes: List = [
+ class_sizes = [
  self.class_sizes[sample_class],
  self.class_sizes[neigh_class],
  ]
@@ -388,12 +388,12 @@ def _get_coefficients(self, gaussian_mixture: GaussianMixture) -> Tuple[np.ndarr
  if self.covariance_type == "tied":
  covariances = np.array([covariances] * gaussian_mixture.n_components)
  elif self.covariance_type == "diag":
- cov_list: List = []
+ cov_list = []
  for component in range(gaussian_mixture.n_components):
  cov_list.append(np.diagflat(covariances[component, :]))
  covariances = np.array(cov_list)
  elif self.covariance_type == "spherical":
- cov_list: List = []
+ cov_list = []
  for component in range(gaussian_mixture.n_components):
  var = np.array([covariances[component]] * self.__x_subset.shape[1])
  cov_list.append(np.diagflat(var))
@@ -431,7 +431,7 @@ def _compute_mdist(self, in_data: np.ndarray, mean: np.ndarray, covariance: np.n
  mdist = cdist(data, [mean], metric="mahalanobis", VI=np.linalg.inv(covariance))[:, 0]
  except Exception as e:
  self.__logger.error("Can't compute 'cdist' function. Distance threshold is set to 2.0")
- self.__logger.info(f"For more information, examine exception: {e}")
+ self.__logger.error(f"For more information, examine exception: {e}")
  mdist = np.full_like(in_data, fill_value=2.0)[:, 0]
  return mdist
 

pyproject.toml

@@ -28,6 +28,7 @@ dependencies = [
  "matplotlib~=3.6.2",
  "click~=8.1.3",
  "scikit-posthocs~=0.7.0",
+ "pydantic~=1.10.5",
 ]
 
 [project.optional-dependencies]

tests/resampling/test_gmm_sampler.py

@@ -1,6 +1,7 @@
+import logging
 import numpy as np
 import pytest
-from collections import Counter
+from collections import Counter, OrderedDict
 
 from multi_imbalance.resampling.gmm_sampler import GMMSampler
 
@@ -45,8 +46,8 @@
 
 @pytest.fixture()
 def gmm_sampler_mock():
- def _get_parametrized_gmm_sampler(X, y, undersample):
- gmm_sampler = GMMSampler(undersample=undersample)
+ def _get_parametrized_gmm_sampler(*args, **kwargs) -> GMMSampler:
+ gmm_sampler = GMMSampler(*args, **kwargs)
  return gmm_sampler
 
  return _get_parametrized_gmm_sampler
@@ -57,13 +58,14 @@ def get_goal_quantity(y):
  return np.mean((quantities[minority_class], quantities[majority_class]), dtype=int)
 
 
+@pytest.mark.parametrize("strategy, filter_new", [("median", 1.0), ("average", 0.0)])
 @pytest.mark.parametrize("X, y", complete_test_data)
-def test_output_length_with_undersample(X, y, gmm_sampler_mock):
- gmm_sampler = gmm_sampler_mock(X, y, True)
+def test_output_length_with_undersample(X, y, strategy, filter_new, gmm_sampler_mock):
+ gmm_sampler = gmm_sampler_mock(undersample=True, strategy=strategy, filter_new=filter_new)
  resampled_X, resampled_y = gmm_sampler.fit_resample(X, y)
 
  y_resampled_count = Counter(resampled_y)
- for _, quantity in y_resampled_count.items():
+ for quantity in y_resampled_count.values():
  assert quantity == get_goal_quantity(y)
 
  assert len(resampled_X) == get_goal_quantity(y) * num_classes
@@ -72,8 +74,8 @@ def test_output_length_with_undersample(X, y, gmm_sampler_mock):
 
 @pytest.mark.parametrize("X, y", complete_test_data)
 def test_output_length_without_undersample(X, y, gmm_sampler_mock):
- gmm_sampler = gmm_sampler_mock(X, y, False)
- resampled_X, resampled_y = gmm_sampler.fit_resample(X, y)
+ gmm_sampler = gmm_sampler_mock(undersample=False)
+ _, resampled_y = gmm_sampler.fit_resample(X, y)
 
  y_count = Counter(y)
  y_resampled_count = Counter(resampled_y)
@@ -83,9 +85,88 @@ def test_output_length_without_undersample(X, y, gmm_sampler_mock):
 
 
 def test_perform_step_condition(gmm_sampler_mock):
- gmm_sampler = GMMSampler()
+ gmm_sampler = gmm_sampler_mock()
  assert gmm_sampler._perform_step(n_components=2, likelihood=1.0, num_samples=3)
  assert not gmm_sampler._perform_step(n_components=2, likelihood=-1.0, num_samples=3)
  assert not gmm_sampler._perform_step(n_components=2, likelihood=1.0, num_samples=1)
  gmm_sampler.max_components = 1
  assert not gmm_sampler._perform_step(n_components=4, likelihood=1.0, num_samples=3)
+
+
+def test_minority_classes(gmm_sampler_mock):
+ minority_classes = [0, 1]
+ gmm_sampler = gmm_sampler_mock(minority_classes=minority_classes)
+
+ assert gmm_sampler.minority_classes == minority_classes
+
+
+@pytest.mark.parametrize(
+ "maj_int_min, expected_size",
+ [
+ ({"maj": [], "int": [], "min": [1]}, 6),
+ ({"maj": [1], "int": [], "min": []}, 6),
+ ({"maj": [], "int": [1], "min": []}, 6),
+ ],
+)
+def test_set_size_to_align(gmm_sampler_mock, expected_size, maj_int_min):
+ gmm_sampler = gmm_sampler_mock()
+ gmm_sampler.class_sizes = Counter(y_imb_hard)
+ gmm_sampler.maj_int_min = OrderedDict(maj_int_min)
+
+ gmm_sampler._set_size_to_align()
+ assert gmm_sampler.size_to_align == expected_size
+
+
+def test_compute_mdist(gmm_sampler_mock, caplog):
+ caplog.set_level(logging.ERROR)
+ gmm_sampler = gmm_sampler_mock()
+ mean = [0, 0]
+ covariance = np.eye(2)
+ x = np.random.multivariate_normal(mean, covariance, size=2)
+
+ gmm_sampler._compute_mdist(x, mean, np.ones((2, 2)))
+
+ no_exception_check = 0
+ for record in caplog.records:
+ if record.levelname == "ERROR":
+ msg = record.message
+ no_exception_check += (
+ msg == "Can't compute 'cdist' function. Distance threshold is set to 2.0"
+ or msg == "For more information, examine exception: Singular matrix"
+ )
+
+ assert no_exception_check == 2
+
+
+@pytest.mark.parametrize(
+ "strategy, class_count, expected_middle_size",
+ [("median", [10, 6, 4], 6), ("median", [4, 12, 4], 4), ("average", [4, 4, 12], 6), ("average", [8, 4, 8], 6)],
+)
+def test_get_middle_size_based_on_strategy(strategy, class_count, expected_middle_size, gmm_sampler_mock):
+ gmm_sampler = gmm_sampler_mock(strategy=strategy)
+
+ gmm_sampler._fit(X, np.array([*[0] * class_count[0], *[1] * class_count[1], *[2] * class_count[2]]))
+ middle_size = gmm_sampler._get_middle_size_based_on_strategy()
+ assert middle_size == expected_middle_size
+
+
+def test_get_middle_size_based_on_strategy_exception(gmm_sampler_mock):
+ gmm_sampler = gmm_sampler_mock()
+ gmm_sampler.strategy = "min"
+
+ with pytest.raises(ValueError) as ex:
+ gmm_sampler._get_middle_size_based_on_strategy()
+
+ assert str(ex.value) == 'Unrecognized min. Only "median" and "average" are allowed.'
+
+
+def test_set_size_to_align_exception(gmm_sampler_mock):
+ maj_int_min = {"maj": [], "int": [], "min": []}
+ gmm_sampler = gmm_sampler_mock()
+ gmm_sampler.class_sizes = Counter(y_imb_hard)
+ gmm_sampler.maj_int_min = OrderedDict(maj_int_min)
+
+ with pytest.raises(ValueError) as ex:
+ gmm_sampler._set_size_to_align()
+
+ assert str(ex.value) == "Bad input - can not obtain desire size."