Skip to content

GMM

Module: gmm.py

This module provides a Gaussian Mixture Model (GMM) clustering implementation using scikit-learn's GaussianMixture.

It optionally allows a user-defined number of components or automatically scans [2..max_k] for the best silhouette score.

Classes:

Name Description
Gmm

Gaussian Mixture Model clustering with optional user-specified n_components or silhouette-based auto selection.
Dependencies
  • numpy
  • sklearn.mixture.GaussianMixture
  • sklearn.metrics.silhouette_score
  • base.py (Clustering)
Key Features
  • Automatic scanning of [2..max_k] for best silhouette score if n_components is None
  • Final model is stored, along with predicted cluster labels
  • Ties into the base Clustering for plotting/metrics
Authors
  • Debajyoti Sahoo (debajyotis@iisc.ac.in)
Version Info
  • 28/Dec/2024: Initial release

Gmm

Bases: Clustering

Gaussian Mixture Model clustering with optional user-defined 'n_components' or automatic silhouette-based selection.

Attributes:

Name Type Description
n_components Optional[int]

The actual number of components used in the final fitted model. If provided, the class will skip auto-selection and directly use this many mixture components.
max_k int

Maximum number of components to consider for auto selection if n_components is None.
labels Optional[ndarray]

Cluster/component labels for each data point after fitting.
Source code in scirex/core/ml/unsupervised/clustering/gmm.py 
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
class Gmm(Clustering):
    """
    Gaussian Mixture Model clustering with optional user-defined 'n_components'
    or automatic silhouette-based selection.

    Attributes:
        n_components (Optional[int]):
            The actual number of components used in the final fitted model.
            If provided, the class will skip auto-selection
            and directly use this many mixture components.
        max_k (int):
            Maximum number of components to consider for auto selection if n_components is None.
        labels (Optional[np.ndarray]):
            Cluster/component labels for each data point after fitting.
    """

    def __init__(self, n_components: Optional[int] = None, max_k: int = 10) -> None:
        """
        Initialize the Gmm clustering class.

        Args:
            n_components (Optional[int], optional):
                If provided, the model will directly use this many Gaussian components.
                Otherwise, it scans [2..max_k] for the best silhouette score. Defaults to None.
            max_k (int, optional):
                Maximum components to try for auto selection if n_components is None. Defaults to 10.
        """
        super().__init__("gmm")
        self.n_components = n_components
        self.max_k = max_k

        # Populated after fitting
        self.labels: Optional[np.ndarray] = None
        self.n_components_: Optional[int] = None
        self.model: Optional[GaussianMixture] = None

    def fit(self, X: np.ndarray) -> None:
        """
        Fit the GMM model to the data.

        If user-defined n_components is set, skip auto selection.
        Otherwise, compute silhouette scores across [2..max_k]
        and pick the best.

        Args:
            X (np.ndarray): Scaled feature matrix of shape (n_samples, n_features).
        """
        X = X.astype(np.float32, copy=False)
        n_samples, n_features = X.shape

        if self.n_components is not None:
            # Use user-specified
            self.n_components_ = self.n_components
            print(f"Fitting GMM with user-defined n_components={self.n_components_}.\n")
        else:
            # Automatic silhouette-based selection
            k_values = range(2, self.max_k + 1)
            silhouettes = []

            # Subsampling for silhouette
            rng = np.random.default_rng(self.random_state)
            sample_size = min(1000, n_samples)

            for k in k_values:
                gmm = GaussianMixture(n_components=k, random_state=self.random_state)
                gmm.fit(X)
                labels_candidate = gmm.predict(X)

                # Must have at least 2 distinct clusters for silhouette
                if len(np.unique(labels_candidate)) > 1:
                    if n_samples > sample_size:
                        indices = rng.choice(n_samples, sample_size, replace=False)
                        X_sample = X[indices]
                        labels_sample = labels_candidate[indices]
                    else:
                        X_sample = X
                        labels_sample = labels_candidate
                    score = silhouette_score(X_sample, labels_sample)
                else:
                    score = -1  # invalid silhouette

                silhouettes.append(score)

            best_k = k_values[np.argmax(silhouettes)]
            self.n_components_ = best_k
            print(f"Optimal k (silhouette) = {best_k}\n")

        self.model = GaussianMixture(
            n_components=self.n_components_, random_state=self.random_state
        )
        self.labels = self.model.fit_predict(X)

        print(f"GMM fitted with n_components={self.n_components_}.\n")

    def get_model_params(self) -> Dict[str, Any]:
        """
        Get parameters/results of the fitted GMM model.

        Returns:
            Dict[str, Any]:
                - n_components (int): The final number of components used
                - max_k (int): The maximum considered if auto
        """
        return {"n_components": self.n_components_, "max_k": self.max_k}

__init__(n_components=None, max_k=10)

Initialize the Gmm clustering class.

Parameters:

Name Type Description Default
n_components Optional[int]

If provided, the model will directly use this many Gaussian components. Otherwise, it scans [2..max_k] for the best silhouette score. Defaults to None.

 None
max_k int

Maximum components to try for auto selection if n_components is None. Defaults to 10.

 10
Source code in scirex/core/ml/unsupervised/clustering/gmm.py 
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
def __init__(self, n_components: Optional[int] = None, max_k: int = 10) -> None:
    """
    Initialize the Gmm clustering class.

    Args:
        n_components (Optional[int], optional):
            If provided, the model will directly use this many Gaussian components.
            Otherwise, it scans [2..max_k] for the best silhouette score. Defaults to None.
        max_k (int, optional):
            Maximum components to try for auto selection if n_components is None. Defaults to 10.
    """
    super().__init__("gmm")
    self.n_components = n_components
    self.max_k = max_k

    # Populated after fitting
    self.labels: Optional[np.ndarray] = None
    self.n_components_: Optional[int] = None
    self.model: Optional[GaussianMixture] = None

fit(X)

Fit the GMM model to the data.

If user-defined n_components is set, skip auto selection. Otherwise, compute silhouette scores across [2..max_k] and pick the best.

Parameters:

Name Type Description Default
X ndarray

Scaled feature matrix of shape (n_samples, n_features).

 required
Source code in scirex/core/ml/unsupervised/clustering/gmm.py 
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
def fit(self, X: np.ndarray) -> None:
    """
    Fit the GMM model to the data.

    If user-defined n_components is set, skip auto selection.
    Otherwise, compute silhouette scores across [2..max_k]
    and pick the best.

    Args:
        X (np.ndarray): Scaled feature matrix of shape (n_samples, n_features).
    """
    X = X.astype(np.float32, copy=False)
    n_samples, n_features = X.shape

    if self.n_components is not None:
        # Use user-specified
        self.n_components_ = self.n_components
        print(f"Fitting GMM with user-defined n_components={self.n_components_}.\n")
    else:
        # Automatic silhouette-based selection
        k_values = range(2, self.max_k + 1)
        silhouettes = []

        # Subsampling for silhouette
        rng = np.random.default_rng(self.random_state)
        sample_size = min(1000, n_samples)

        for k in k_values:
            gmm = GaussianMixture(n_components=k, random_state=self.random_state)
            gmm.fit(X)
            labels_candidate = gmm.predict(X)

            # Must have at least 2 distinct clusters for silhouette
            if len(np.unique(labels_candidate)) > 1:
                if n_samples > sample_size:
                    indices = rng.choice(n_samples, sample_size, replace=False)
                    X_sample = X[indices]
                    labels_sample = labels_candidate[indices]
                else:
                    X_sample = X
                    labels_sample = labels_candidate
                score = silhouette_score(X_sample, labels_sample)
            else:
                score = -1  # invalid silhouette

            silhouettes.append(score)

        best_k = k_values[np.argmax(silhouettes)]
        self.n_components_ = best_k
        print(f"Optimal k (silhouette) = {best_k}\n")

    self.model = GaussianMixture(
        n_components=self.n_components_, random_state=self.random_state
    )
    self.labels = self.model.fit_predict(X)

    print(f"GMM fitted with n_components={self.n_components_}.\n")

get_model_params()

Get parameters/results of the fitted GMM model.

Returns:

Type Description
Dict[str, Any]

Dict[str, Any]: - n_components (int): The final number of components used - max_k (int): The maximum considered if auto
Source code in scirex/core/ml/unsupervised/clustering/gmm.py 
162
163
164
165
166
167
168
169
170
171
def get_model_params(self) -> Dict[str, Any]:
    """
    Get parameters/results of the fitted GMM model.

    Returns:
        Dict[str, Any]:
            - n_components (int): The final number of components used
            - max_k (int): The maximum considered if auto
    """
    return {"n_components": self.n_components_, "max_k": self.max_k}