update docstrings

multi_imbalance/resampling/gmm_sampler.py

@@ -22,34 +22,45 @@ class GMMSampler(BaseSampler):
  """
  GGMSampling algorithm that uses creating new examples by sampling from a multivariate normal distribution
  (whose parameters are estimated from the input data) and removing troublesome examples from the majority class.
+
  Parameters
  ----------
  likelihood_threshold : float, default=0.0
  Minimum likelihood change threshold. A value below this threshold will be equivalent to no change.
+
  k_neighbors : int, default=7
  The number of analyzed nearest neighbors during the analysis.
  Used during both undersampling and oversampling.
+
  undersample : bool, default=True
  A binary value indicating whether to perform an undersampling operation on majority classes.
+
  min_components : int, default=1
  Minimum number of components of GaussianMixture.
+
  max_components : Optional[int], default=None
  Maximum number of components of GaussianMixture. Without upper bound if not specified.
+
  minority_classes : Optional[List[int]], default=None
  List containing minority classes given by hand - no auto detection of minority classes will be done.
+
  valid_size : float, default=0.25
  Size of validation set to perform test for components choosing.
+
  filter_new : float, default=-1
  Parameter controlling the behavior after the oversampling operation.
  Checks if and how to filter newly created examples:
  -1 -> do not filter out
  0 -> filter out by max/mean value of created examples
  >0 -> specify your own value e.g. 2.0
+
  add_after_filtration : bool, default=True
  Value specifying whether to regenerate the examples after filtering.
+
  iterations_after_filtration : int, default=50
  This value will potentially avoid an endless loop of deleting and re-generating examples.
  The upper limit for the number of repetitions.
+
  covariance_type : "full", "tied", "diag", "spherical", default="full"
  String describing the type of covariance parameters to use in GaussianMixture. Must be one of:
  - "full"
@@ -60,35 +71,48 @@ class GMMSampler(BaseSampler):
  each component has its own diagonal covariance matrix
  - "spherical"
  each component has its own single variance
+
  strategy : str "average" or "median", default="average"
  The strategy of selecting the number of examples considers the target number in each class.
+
  {random_state}
+
  n_init : int, default=10
  The number of initializations to perform in GaussianMixture. The best results are kept.
+
  tol : float, default=1e-3
  The convergence threshold in GaussianMixture. EM iterations will stop when the lower bound
  average gain is below this threshold.
+
  max_iter : int, default=100
  The number of EM iterations to perform in GaussianMixture.
+
  Attributes
  ----------
  likelihoods : dict
  Likelihood of each minority class obtained after fitting the final Gaussian model.
+
  gaussian_mixtures : dict
  Dictionary containing all Gaussian models for each minority class separately.
+
  class_sizes : Optional[Counter]
  A dictionary containing the counts of each class.
+
  neighborhood : Optional[dict]
  Dictionary with information on the nearest points for each example separately.
+
  maj_int_min : OrderedDict
  A dictionary containing information on which class can be considered majority,
  which minority and which remaining class - a heuristic matching.
+
  size_to_align : Optional[np.ndarray]
  ndarray containing information about the quantity considered the gold standard -
  it is this size that we will want to generate and remove examples.
+
  cdist_min_count : int
  The minimum number of examples found in the data sample on which
  distances between points are calculated (by the cdist method).
+
  Examples
  --------
  >>> import numpy as np