Skip to content

Subspace Smooth Matching Pursuit

Solver ID: SubSMP

Usage

from invert import Solver

# fwd = ...    (mne.Forward object)
# evoked = ... (mne.Evoked object)

solver = Solver("SubSMP")
solver.make_inverse_operator(fwd)
stc = solver.apply_inverse_operator(evoked)
stc.plot()

Overview

Smooth matching pursuit variant that projects data into a low-dimensional subspace and merges patch supports across components.

References

  1. Lukas Hecker 2025, unpublished

API Reference

Bases: BaseSolver

Class for the Subspace Smooth Matching Pursuit (SubSMP) inverse solution. Developed by Lukas Hecker as a smooth extension of the orthogonal matching pursuit algorithm [1], 19.10.2022.

References

[1] Duarte, M. F., & Eldar, Y. C. (2011). Structured compressed sensing: From theory to applications. IEEE Transactions on signal processing, 59(9), 4053-4085.

Source code in invert/solvers/matching_pursuit/subsmp.py 
 18
 19
 20
 21
 22
 23
 24
 25
 26
 27
 28
 29
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 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
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
class SolverSubSMP(BaseSolver):
    """Class for the Subspace Smooth Matching Pursuit (SubSMP) inverse solution. Developed
        by Lukas Hecker as a smooth extension of the orthogonal matching pursuit
        algorithm [1], 19.10.2022.


    References
    ----------
    [1] Duarte, M. F., & Eldar, Y. C. (2011). Structured compressed sensing:
    From theory to applications. IEEE Transactions on signal processing, 59(9),
    4053-4085.

    """

    meta = SolverMeta(
        acronym="SubSMP",
        full_name="Subspace Smooth Matching Pursuit",
        category="Matching Pursuit",
        description=(
            "Smooth matching pursuit variant that projects data into a low-dimensional "
            "subspace and merges patch supports across components."
        ),
        references=[
            "Lukas Hecker 2025, unpublished",
        ],
    )

    def __init__(self, name="Subspace Smooth Matching Pursuit", **kwargs):
        self.name = name
        return super().__init__(**kwargs)

    def make_inverse_operator(self, forward, *args, alpha="auto", **kwargs):
        """Calculate inverse operator.

        Parameters
        ----------
        forward : mne.Forward
            The mne-python Forward model instance.
        alpha : float
            The regularization parameter.

        Return
        ------
        self : object returns itself for convenience
        """
        super().make_inverse_operator(forward, *args, alpha=alpha, **kwargs)
        self.inverse_operators = []

        adjacency = mne.spatial_src_adjacency(self.forward["src"], verbose=0).toarray()
        self.adjacency = adjacency
        laplace_operator = laplacian(adjacency)
        self.laplace_operator = laplace_operator

        patch_operator = adjacency + np.eye(adjacency.shape[0])
        leadfield_smooth = self.leadfield @ patch_operator

        self.leadfield_smooth = leadfield_smooth
        norms = np.linalg.norm(self.leadfield_smooth, axis=0)
        norms[norms == 0] = 1
        self.leadfield_smooth_normed = self.leadfield_smooth / norms
        norms = np.linalg.norm(self.leadfield, axis=0)
        norms[norms == 0] = 1
        self.leadfield_normed = self.leadfield / norms

        return self

    def apply_inverse_operator(
        self, mne_obj, include_singletons=True
    ) -> mne.SourceEstimate:
        """Apply the inverse operator.
        Parameters
        ----------
        mne_obj : [mne.Evoked, mne.Epochs, mne.io.Raw]
            The MNE data object.

        Return
        ------
        stc : mne.SourceEstimate
            The mne Source Estimate object
        """
        data = self.unpack_data_obj(mne_obj)
        source_mat = self.calc_subsmp_solution(
            data, include_singletons=include_singletons
        )
        stc = self.source_to_object(source_mat)
        return stc

    def calc_subsmp_solution(self, y, include_singletons=True, var_thresh=1):
        """Calculate the Subspace Smooth Matching Pursuit (SubSMP) solution.

        Parameters
        ----------
        y : numpy.ndarray
            The M/EEG data Matrix
        include_singletons : bool
            If True -> include single dipoles as candidates,
            else include only smooth patches
        var_thresh : float
            Threshold how much variance will be explained in the data

        """
        _, n_time = y.shape
        n_dipoles = self.leadfield.shape[1]

        u, s, _ = np.linalg.svd(y, full_matrices=False)

        # Select components using base class method (find_corner returns an index, need count)
        n_comps = max(self.get_comps_L(s) + 1, 1)
        logger.info(f"Selected {n_comps} components")

        # Find matching sources to each of the selected eigenvectors (components)
        topos = u[:, :n_comps]
        x_hats = []
        for topo in topos.T:
            x_hats.append(
                self.calc_smp_solution(
                    topo, include_singletons=include_singletons, var_thresh=var_thresh
                )
            )

        # Extract all active dipoles and patches:
        omegas = np.unique(
            np.where((np.stack(x_hats, axis=0) ** 2).sum(axis=0) != 0)[0]
        )

        # Calculate minimum-norm solution at active dipoles and patches
        x_hat = np.zeros((n_dipoles, n_time))

        x_hat[omegas, :], _, _, _ = np.linalg.lstsq(
            self.leadfield[:, omegas], y, rcond=None
        )

        return x_hat

    def calc_smp_solution(self, y, include_singletons=True, var_thresh=1):
        """Calculates the Smooth Matching Pursuit (SMP) inverse solution.

        Parameters
        ----------
        y : numpy.ndarray
            The data matrix (channels,).
        include_singletons : bool
            If True -> Include not only smooth patches but also single dipoles.
        var_thresh : float
            Threshold how much variance will be explained in the data

        Return
        ------
        x_hat : numpy.ndarray
            The inverse solution (dipoles,)
        """
        n_chans = len(y)
        _, n_dipoles = self.leadfield.shape

        x_hat = np.zeros(n_dipoles)
        omega = np.array([], dtype=int)
        r = deepcopy(y)
        y_hat = self.leadfield @ x_hat
        residuals = np.array(
            [
                np.linalg.norm(y - y_hat),
            ]
        )
        unexplained_variance = np.array(
            [
                calc_residual_variance(y_hat, y),
            ]
        )
        x_hats = [
            deepcopy(x_hat),
        ]

        for _ in range(n_chans):
            b_smooth = self.leadfield_smooth_normed.T @ r
            b_sparse = self.leadfield_normed.T @ r

            if include_singletons and (abs(b_sparse).max() > abs(b_smooth).max()):
                b_sparse_thresh = thresholding(b_sparse, 1)
                new_patch = np.where(b_sparse_thresh != 0)[0]
            else:
                b_smooth_thresh = thresholding(b_smooth, 1)
                best_idx = np.where(b_smooth_thresh != 0)[0]
                new_patch = np.where(self.adjacency[best_idx[0]] != 0)[0]
                new_patch = np.append(new_patch, best_idx)

            omega = np.unique(np.append(omega, new_patch)).astype(int)
            x_hat = np.zeros(n_dipoles)
            x_hat[omega], _, _, _ = np.linalg.lstsq(
                self.leadfield[:, omega], y, rcond=None
            )
            y_hat = self.leadfield @ x_hat
            r = y - y_hat

            residuals = np.append(residuals, np.linalg.norm(r))
            unexplained_variance = np.append(
                unexplained_variance, calc_residual_variance(y_hat, y)
            )
            x_hats.append(deepcopy(x_hat))
            if residuals[-1] > residuals[-2]:
                break

        if unexplained_variance[1] > var_thresh:
            x_hat = best_index_residual(unexplained_variance, x_hats, plot=False)
        else:
            x_hat = x_hats[1]

        return x_hat

__init__ 

__init__(name='Subspace Smooth Matching Pursuit', **kwargs)
Source code in invert/solvers/matching_pursuit/subsmp.py 
45
46
47
def __init__(self, name="Subspace Smooth Matching Pursuit", **kwargs):
    self.name = name
    return super().__init__(**kwargs)

make_inverse_operator 

make_inverse_operator(
    forward, *args, alpha="auto", **kwargs
)

Calculate inverse operator.

Parameters:

Name Type Description Default
forward Forward

The mne-python Forward model instance.

 required
alpha float

The regularization parameter.

 'auto'
Return

self : object returns itself for convenience

Source code in invert/solvers/matching_pursuit/subsmp.py 
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
def make_inverse_operator(self, forward, *args, alpha="auto", **kwargs):
    """Calculate inverse operator.

    Parameters
    ----------
    forward : mne.Forward
        The mne-python Forward model instance.
    alpha : float
        The regularization parameter.

    Return
    ------
    self : object returns itself for convenience
    """
    super().make_inverse_operator(forward, *args, alpha=alpha, **kwargs)
    self.inverse_operators = []

    adjacency = mne.spatial_src_adjacency(self.forward["src"], verbose=0).toarray()
    self.adjacency = adjacency
    laplace_operator = laplacian(adjacency)
    self.laplace_operator = laplace_operator

    patch_operator = adjacency + np.eye(adjacency.shape[0])
    leadfield_smooth = self.leadfield @ patch_operator

    self.leadfield_smooth = leadfield_smooth
    norms = np.linalg.norm(self.leadfield_smooth, axis=0)
    norms[norms == 0] = 1
    self.leadfield_smooth_normed = self.leadfield_smooth / norms
    norms = np.linalg.norm(self.leadfield, axis=0)
    norms[norms == 0] = 1
    self.leadfield_normed = self.leadfield / norms

    return self

apply_inverse_operator 

apply_inverse_operator(
    mne_obj, include_singletons=True
) -> mne.SourceEstimate

Apply the inverse operator.

Parameters:

Name Type Description Default
mne_obj [Evoked, Epochs, Raw]

The MNE data object.

 required
Return

stc : mne.SourceEstimate The mne Source Estimate object

Source code in invert/solvers/matching_pursuit/subsmp.py 
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
def apply_inverse_operator(
    self, mne_obj, include_singletons=True
) -> mne.SourceEstimate:
    """Apply the inverse operator.
    Parameters
    ----------
    mne_obj : [mne.Evoked, mne.Epochs, mne.io.Raw]
        The MNE data object.

    Return
    ------
    stc : mne.SourceEstimate
        The mne Source Estimate object
    """
    data = self.unpack_data_obj(mne_obj)
    source_mat = self.calc_subsmp_solution(
        data, include_singletons=include_singletons
    )
    stc = self.source_to_object(source_mat)
    return stc