Source code for gtda.homology.simplicial

"""Persistent homology on point clouds or finite metric spaces."""
# License: GNU AGPLv3

from numbers import Real, Integral
from types import FunctionType

import numpy as np
from joblib import Parallel, delayed
from pyflagser import flagser_weighted
from scipy.sparse import coo_matrix
from scipy.spatial import Delaunay
from sklearn.base import BaseEstimator, TransformerMixin
from sklearn.metrics.pairwise import pairwise_distances
from sklearn.utils.validation import check_is_fitted

from ._utils import _postprocess_diagrams
from ..base import PlotterMixin
from ..externals.python import ripser, SparseRipsComplex, CechComplex
from ..plotting import plot_diagram
from ..utils._docs import adapt_fit_transform_docs
from ..utils.intervals import Interval
from ..utils.validation import validate_params, check_point_clouds

_AVAILABLE_RIPS_WEIGHTS = {
    "DTM": {
        "p": {"type": Real, "in": [1, 2, np.inf]},
        "r": {"type": Real, "in": Interval(0, np.inf, closed="right")},
        "n_neighbors": {"type": Integral,
                        "in": Interval(1, np.inf, closed="left")}
        },
    "general": {
        "p": {"type": Real, "in": [1, 2, np.inf]},
        }
    }


[docs]@adapt_fit_transform_docs class VietorisRipsPersistence(BaseEstimator, TransformerMixin, PlotterMixin): """:ref:`Persistence diagrams <persistence_diagram>` resulting from :ref:`Vietoris–Rips filtrations <vietoris-rips_complex_and_vietoris-rips_persistence>`. Given a :ref:`point cloud <distance_matrices_and_point_clouds>` in Euclidean space, an abstract :ref:`metric space <distance_matrices_and_point_clouds>` encoded by a distance matrix, or the adjacency matrix of a weighted undirected graph, information about the appearance and disappearance of topological features (technically, :ref:`homology classes <homology_and_cohomology>`) of various dimensions and at different scales is summarised in the corresponding persistence diagram. **Important note**: - Persistence diagrams produced by this class must be interpreted with care due to the presence of padding triples which carry no information. See :meth:`transform` for additional information. Parameters ---------- metric : string or callable, optional, default: ``"euclidean"`` If set to ``"precomputed"``, input data is to be interpreted as a collection of distance matrices or of adjacency matrices of weighted undirected graphs. Otherwise, input data is to be interpreted as a collection of point clouds (i.e. feature arrays), and `metric` determines a rule with which to calculate distances between pairs of points (i.e. row vectors). If `metric` is a string, it must be one of the options allowed by :func:`scipy.spatial.distance.pdist` for its metric parameter, or a metric listed in :obj:`sklearn.pairwise.PAIRWISE_DISTANCE_FUNCTIONS`, including ``"euclidean"``, ``"manhattan"`` or ``"cosine"``. If `metric` is a callable, it should take pairs of vectors (1D arrays) as input and, for each two vectors in a pair, it should return a scalar indicating the distance/dissimilarity between them. metric_params : dict, optional, default: ``{}`` Additional parameters to be passed to the distance function. homology_dimensions : list or tuple, optional, default: ``(0, 1)`` Dimensions (non-negative integers) of the topological features to be detected. coeff : int prime, optional, default: ``2`` Compute homology with coefficients in the prime field :math:`\\mathbb{F}_p = \\{ 0, \\ldots, p - 1 \\}` where :math:`p` equals `coeff`. collapse_edges : bool, optional, default: ``False`` Whether to run the edge collapse algorithm in [2]_ prior to the persistent homology computation (see the Notes). Can reduce the runtime dramatically when the data or the maximum homology dimensions are large. max_edge_length : float, optional, default: ``numpy.inf`` Maximum value of the Vietoris–Rips filtration parameter. Points whose distance is greater than this value will never be connected by an edge, and topological features at scales larger than this value will not be detected. infinity_values : float or None, default: ``None`` Which death value to assign to features which are still alive at filtration value `max_edge_length`. ``None`` means that this death value is declared to be equal to `max_edge_length`. reduced_homology : bool, optional, default: ``True`` If ``True``, the earliest-born triple in homology dimension 0 which has infinite death is discarded from each diagram computed in :meth:`transform`. n_jobs : int or None, optional, default: ``None`` The number of jobs to use for the computation. ``None`` means 1 unless in a :obj:`joblib.parallel_backend` context. ``-1`` means using all processors. Attributes ---------- infinity_values_ : float Effective death value to assign to features which are still alive at filtration value `max_edge_length`. See also -------- WeightedRipsPersistence, FlagserPersistence, SparseRipsPersistence, WeakAlphaPersistence, EuclideanCechPersistence, ConsistentRescaling, ConsecutiveRescaling Notes ----- `Ripser <https://github.com/Ripser/ripser>`_ [1]_ is used as a C++ backend for computing Vietoris–Rips persistent homology. Python bindings were modified for performance from the `ripser.py <https://github.com/scikit-tda/ripser.py>`_ package. `GUDHI <https://github.com/GUDHI/gudhi-devel>`_ is used as a C++ backend for the edge collapse algorithm described in [2]_. References ---------- .. [1] U. Bauer, "Ripser: efficient computation of Vietoris–Rips persistence barcodes", 2019; `arXiv:1908.02518 <https://arxiv.org/abs/1908.02518>`_. .. [2] J.-D. Boissonnat and S. Pritam, "Edge Collapse and Persistence of Flag Complexes"; in *36th International Symposium on Computational Geometry (SoCG 2020)*, pp. 19:1–19:15, Schloss Dagstuhl-Leibniz–Zentrum für Informatik, 2020; `DOI: 10.4230/LIPIcs.SoCG.2020.19 <https://doi.org/10.4230/LIPIcs.SoCG.2020.19>`_. """ _hyperparameters = { "metric": {"type": (str, FunctionType)}, "metric_params": {"type": dict}, "homology_dimensions": { "type": (list, tuple), "of": {"type": int, "in": Interval(0, np.inf, closed="left")} }, "collapse_edges": {"type": bool}, "coeff": {"type": int, "in": Interval(2, np.inf, closed="left")}, "max_edge_length": {"type": Real}, "infinity_values": {"type": (Real, type(None))}, "reduced_homology": {"type": bool} }
[docs] def __init__(self, metric="euclidean", metric_params={}, homology_dimensions=(0, 1), collapse_edges=False, coeff=2, max_edge_length=np.inf, infinity_values=None, reduced_homology=True, n_jobs=None): self.metric = metric self.metric_params = metric_params self.homology_dimensions = homology_dimensions self.collapse_edges = collapse_edges self.coeff = coeff self.max_edge_length = max_edge_length self.infinity_values = infinity_values self.reduced_homology = reduced_homology self.n_jobs = n_jobs
def _ripser_diagram(self, X): Xdgms = ripser( X, maxdim=self._max_homology_dimension, thresh=self.max_edge_length, coeff=self.coeff, metric=self.metric, metric_params=self.metric_params, collapse_edges=self.collapse_edges )["dgms"] return Xdgms
[docs] def fit(self, X, y=None): """Calculate :attr:`infinity_values_`. Then, return the estimator. This method is here to implement the usual scikit-learn API and hence work in pipelines. Parameters ---------- X : ndarray or list of length n_samples Input data representing a collection of point clouds if `metric` was not set to ``"precomputed"``, and of distance matrices or adjacency matrices of weighted undirected graphs otherwise. Can be either a 3D ndarray whose zeroth dimension has size ``n_samples``, or a list containing ``n_samples`` 2D ndarrays/sparse matrices. Point cloud arrays have shape ``(n_points, n_dimensions)``, and if `X` is a list these shapes can vary between point clouds. If `metric` was set to ``"precomputed"``, then: - Diagonal entries indicate vertex weights, i.e. the filtration parameters at which vertices appear. - If entries of `X` are dense, only their upper diagonal portions (including the diagonal) are considered. - If entries of `X` are sparse, they do not need to be upper diagonal or symmetric. If only one of entry (i, j) and (j, i) is stored, its value is taken as the weight of the undirected edge {i, j}. If both are stored, the value in the upper diagonal is taken. Off-diagonal entries which are not explicitly stored are treated as infinite, indicating absent edges. - Entries of `X` should be compatible with a filtration, i.e. the value at index (i, j) should be no smaller than the values at diagonal indices (i, i) and (j, j). y : None There is no need for a target in a transformer, yet the pipeline API requires this parameter. Returns ------- self : object """ validate_params( self.get_params(), self._hyperparameters, exclude=["n_jobs"]) self._is_precomputed = self.metric == "precomputed" check_point_clouds(X, accept_sparse=True, distance_matrices=self._is_precomputed) if self.infinity_values is None: self.infinity_values_ = self.max_edge_length else: self.infinity_values_ = self.infinity_values self._homology_dimensions = sorted(self.homology_dimensions) self._max_homology_dimension = self._homology_dimensions[-1] return self
[docs] def transform(self, X, y=None): """For each point cloud or distance matrix in `X`, compute the relevant persistence diagram as an array of triples [b, d, q]. Each triple represents a persistent topological feature in dimension q (belonging to `homology_dimensions`) which is born at b and dies at d. Only triples in which b < d are meaningful. Triples in which b and d are equal ("diagonal elements") may be artificially introduced during the computation for padding purposes, since the number of non-trivial persistent topological features is typically not constant across samples. They carry no information and hence should be effectively ignored by any further computation. Parameters ---------- X : ndarray or list of length n_samples Input data representing a collection of point clouds if `metric` was not set to ``"precomputed"``, and of distance matrices or adjacency matrices of weighted undirected graphs otherwise. Can be either a 3D ndarray whose zeroth dimension has size ``n_samples``, or a list containing ``n_samples`` 2D ndarrays/sparse matrices. Point cloud arrays have shape ``(n_points, n_dimensions)``, and if `X` is a list these shapes can vary between point clouds. If `metric` was set to ``"precomputed"``, then: - Diagonal entries indicate vertex weights, i.e. the filtration parameters at which vertices appear. - If entries of `X` are dense, only their upper diagonal portions (including the diagonal) are considered. - If entries of `X` are sparse, they do not need to be upper diagonal or symmetric. If only one of entry (i, j) and (j, i) is stored, its value is taken as the weight of the undirected edge {i, j}. If both are stored, the value in the upper diagonal is taken. Off-diagonal entries which are not explicitly stored are treated as infinite, indicating absent edges. - Entries of `X` should be compatible with a filtration, i.e. the value at index (i, j) should be no smaller than the values at diagonal indices (i, i) and (j, j). y : None There is no need for a target in a transformer, yet the pipeline API requires this parameter. Returns ------- Xt : ndarray of shape (n_samples, n_features, 3) Array of persistence diagrams computed from the feature arrays or distance matrices in `X`. ``n_features`` equals :math:`\\sum_q n_q`, where :math:`n_q` is the maximum number of topological features in dimension :math:`q` across all samples in `X`. """ check_is_fitted(self) X = check_point_clouds(X, accept_sparse=True, distance_matrices=self._is_precomputed) Xt = Parallel(n_jobs=self.n_jobs)( delayed(self._ripser_diagram)(x) for x in X) Xt = _postprocess_diagrams( Xt, "ripser", self._homology_dimensions, self.infinity_values_, self.reduced_homology ) return Xt
[docs] @staticmethod def plot(Xt, sample=0, homology_dimensions=None, plotly_params=None): """Plot a sample from a collection of persistence diagrams, with homology in multiple dimensions. Parameters ---------- Xt : ndarray of shape (n_samples, n_features, 3) Collection of persistence diagrams, such as returned by :meth:`transform`. sample : int, optional, default: ``0`` Index of the sample in `Xt` to be plotted. homology_dimensions : list, tuple or None, optional, default: ``None`` Which homology dimensions to include in the plot. ``None`` means plotting all dimensions present in ``Xt[sample]``. plotly_params : dict or None, optional, default: ``None`` Custom parameters to configure the plotly figure. Allowed keys are ``"traces"`` and ``"layout"``, and the corresponding values should be dictionaries containing keyword arguments as would be fed to the :meth:`update_traces` and :meth:`update_layout` methods of :class:`plotly.graph_objects.Figure`. Returns ------- fig : :class:`plotly.graph_objects.Figure` object Plotly figure. """ return plot_diagram( Xt[sample], homology_dimensions=homology_dimensions, plotly_params=plotly_params )
[docs]@adapt_fit_transform_docs class WeightedRipsPersistence(BaseEstimator, TransformerMixin, PlotterMixin): """:ref:`Persistence diagrams <persistence_diagram>` resulting from :ref:`weighted Vietoris–Rips filtrations <TODO>` as in [3]_. Given a :ref:`point cloud <distance_matrices_and_point_clouds>` in Euclidean space, an abstract :ref:`metric space <distance_matrices_and_point_clouds>` encoded by a distance matrix, or the adjacency matrix of a weighted undirected graph, information about the appearance and disappearance of topological features (technically, :ref:`homology classes <homology_and_cohomology>`) of various dimensions and at different scales is summarised in the corresponding persistence diagram. Weighted (Vietoris–)Rips filtrations can be useful to highlight topological features against outliers and noise. Among them, the distance-to-measure (DTM) filtration is particularly suited to point clouds due to several favourable properties. This implementation follows the general framework described in [3]_. The idea is that, starting from a way to compute vertex weights :math:`\\{w_i\\}_i` from an input point cloud/distance matrix/adjacency matrix, a modified adjacency matrix is determined whose diagonal entries are the :math:`\\{w_i\\}_i`, and whose edge weights are .. math:: w_{ij} = \\begin{cases} \\max\\{ w_i, w_j \\} &\\text{if } 2\\mathrm{dist}_{ij} \\leq |w_i^p - w_j^p|^{\\frac{1}{p}}, \\\\ t &\\text{otherwise} \\end{cases} where :math:`t` is the only positive root of .. math:: 2 \\mathrm{dist}_{ij} = (t^p - w_i^p)^\\frac{1}{p} + (t^p - w_j^p)^\\frac{1}{p} and :math:`p` is a parameter (see `metric_params`). The modified adjacency matrices are then treated exactly as in :class:`VietorisRipsPersistence`. **Important notes**: - Vertex and edge weights are twice the ones in [3]_ so that the same results as :class:`VietorisRipsPersistence` are obtained when all vertex weights are zero. - Persistence diagrams produced by this class must be interpreted with care due to the presence of padding triples which carry no information. See :meth:`transform` for additional information. Parameters ---------- metric : string or callable, optional, default: ``"euclidean"`` If set to ``"precomputed"``, input data is to be interpreted as a collection of distance matrices or of adjacency matrices of weighted undirected graphs. Otherwise, input data is to be interpreted as a collection of point clouds (i.e. feature arrays), and `metric` determines a rule with which to calculate distances between pairs of points (i.e. row vectors). If `metric` is a string, it must be one of the options allowed by :func:`scipy.spatial.distance.pdist` for its metric parameter, or a metric listed in :obj:`sklearn.pairwise.PAIRWISE_DISTANCE_FUNCTIONS`, including ``"euclidean"``, ``"manhattan"`` or ``"cosine"``. If `metric` is a callable, it should take pairs of vectors (1D arrays) as input and, for each two vectors in a pair, it should return a scalar indicating the distance/dissimilarity between them. metric_params : dict, optional, default: ``{}`` Additional parameters to be passed to the distance function. homology_dimensions : list or tuple, optional, default: ``(0, 1)`` Dimensions (non-negative integers) of the topological features to be detected. weights : ``"DTM"`` or callable, optional, default: ``"DTM"`` Function that will be applied to each input point cloud/distance matrix/adjacency matrix to compute a 1D array of vertex weights for the the modified adjacency matrices. The default ``"DTM"`` denotes the empirical distance-to-measure function defined, following [3]_, by .. math:: w(x) = 2\\left(\\frac{1}{n+1} \\sum_{k=1}^n \\mathrm{dist}(x, x_k)^r\\right)^{1/r}. Here, :math:`\\mathrm{dist}` is the distance metric used, :math:`x_k` is the :math:`k`-th :math:`\\mathrm{dist}`-nearest neighbour of :math:`x` (:math:`x` is not considered a neighbour of itself), :math:`n` is the number of nearest neighbors to include, and :math:`r` is a parameter (see `weight_params`). If a callable, it must return non-negative 1D arrays. weight_params : dict, optional, default: ``{}`` Additional parameters for the weighted filtration. ``"p"`` determines the power to be used in computing edge weights from vertex weights. It can be one of ``1``, ``2`` or ``np.inf`` and defaults to ``1``. If `weights` is ``"DTM"``, the additional keys ``"r"`` (default: ``2``) and ``"n_neighbors"`` (default: ``3``) are available (see `weights`, where the latter corresponds to :math:`n`). coeff : int prime, optional, default: ``2`` Compute homology with coefficients in the prime field :math:`\\mathbb{F}_p = \\{ 0, \\ldots, p - 1 \\}` where :math:`p` equals `coeff`. collapse_edges : bool, optional, default: ``False`` Whether to run the edge collapse algorithm in [2]_ prior to the persistent homology computation (see the Notes). Can reduce the runtime dramatically when the data or the maximum homology dimensions are large. max_edge_weight : float, optional, default: ``numpy.inf`` Maximum value of the filtration parameter in the modified adjacency matrix. Edges with weight greater than this value will be considered absent. infinity_values : float or None, default: ``None`` Which death value to assign to features which are still alive at filtration value `max_edge_weight`. ``None`` means that this death value is declared to be equal to `max_edge_weight`. reduced_homology : bool, optional, default: ``True`` If ``True``, the earliest-born triple in homology dimension 0 which has infinite death is discarded from each diagram computed in :meth:`transform`. n_jobs : int or None, optional, default: ``None`` The number of jobs to use for the computation. ``None`` means 1 unless in a :obj:`joblib.parallel_backend` context. ``-1`` means using all processors. Attributes ---------- infinity_values_ : float Effective death value to assign to features which are still alive at filtration value `max_edge_weight`. effective_weight_params_ : dict Effective parameters involved in computing the weighted Rips filtration. See also -------- VietorisRipsPersistence, SparseRipsPersistence, FlagserPersistence, WeakAlphaPersistence, EuclideanCechPersistence, ConsistentRescaling, ConsecutiveRescaling Notes ----- `Ripser <https://github.com/Ripser/ripser>`_ [1]_ is used as a C++ backend for computing Vietoris–Rips persistent homology. Python bindings were modified for performance from the `ripser.py <https://github.com/scikit-tda/ripser.py>`_ package. `GUDHI <https://github.com/GUDHI/gudhi-devel>`_ is used as a C++ backend for the edge collapse algorithm described in [2]_. References ---------- .. [1] U. Bauer, "Ripser: efficient computation of Vietoris–Rips persistence barcodes", 2019; `arXiv:1908.02518 <https://arxiv.org/abs/1908.02518>`_. .. [2] J.-D. Boissonnat and S. Pritam, "Edge Collapse and Persistence of Flag Complexes"; in *36th International Symposium on Computational Geometry (SoCG 2020)*, pp. 19:1–19:15, Schloss Dagstuhl-Leibniz–Zentrum für Informatik, 2020; `DOI: 10.4230/LIPIcs.SoCG.2020.19 <https://doi.org/10.4230/LIPIcs.SoCG.2020.19>`_. .. [3] H. Anai et al, "DTM-Based Filtrations"; in *Topological Data Analysis* (Abel Symposia, vol 15), Springer, 2020; `DOI: 10.1007/978-3-030-43408-3_2 <https://doi.org/10.1007/978-3-030-43408-3_2>`_. """ _hyperparameters = { "metric": {"type": (str, FunctionType)}, "metric_params": {"type": dict}, "homology_dimensions": { "type": (list, tuple), "of": {"type": int, "in": Interval(0, np.inf, closed="left")} }, "weights": {"type": (str, FunctionType)}, "weight_params": {"type": dict}, "collapse_edges": {"type": bool}, "coeff": {"type": int, "in": Interval(2, np.inf, closed="left")}, "max_edge_weight": {"type": Real}, "infinity_values": {"type": (Real, type(None))}, "reduced_homology": {"type": bool} }
[docs] def __init__(self, metric="euclidean", metric_params={}, homology_dimensions=(0, 1), weights="DTM", weight_params={}, collapse_edges=False, coeff=2, max_edge_weight=np.inf, infinity_values=None, reduced_homology=True, n_jobs=None): self.metric = metric self.metric_params = metric_params self.homology_dimensions = homology_dimensions self.weights = weights self.weight_params = weight_params self.collapse_edges = collapse_edges self.coeff = coeff self.max_edge_weight = max_edge_weight self.infinity_values = infinity_values self.reduced_homology = reduced_homology self.n_jobs = n_jobs
def _ripser_diagram(self, X): if isinstance(self.weights, FunctionType): weights = self.weights(X) else: weights = self.weights Xdgms = ripser( X, maxdim=self._max_homology_dimension, thresh=self.max_edge_weight, coeff=self.coeff, metric=self.metric, metric_params=self.metric_params, weights=weights, weight_params=self.effective_weight_params_, collapse_edges=self.collapse_edges )["dgms"] return Xdgms
[docs] def fit(self, X, y=None): """Calculate :attr:`infinity_values_`. Then, return the estimator. This method is here to implement the usual scikit-learn API and hence work in pipelines. Parameters ---------- X : ndarray or list of length n_samples Input data representing a collection of point clouds if `metric` was not set to ``"precomputed"``, and of distance matrices or adjacency matrices of weighted undirected graphs otherwise. Can be either a 3D ndarray whose zeroth dimension has size ``n_samples``, or a list containing ``n_samples`` 2D ndarrays/sparse matrices. Point cloud arrays have shape ``(n_points, n_dimensions)``, and if `X` is a list these shapes can vary between point clouds. If `metric` was set to ``"precomputed"``, then: - All entries of `X` should not contain infinities or negative values (contrary to :class:`VietorisRipsPersistence`). - The diagonals of entries of `X` are ignored (after the vertex weights are computed, when `weights` is a callable). - If entries of `X` are dense, only their upper diagonal portions are considered. - If entries of `X` are sparse, they do not need to be upper diagonal or symmetric. If only one of entry (i, j) and (j, i) is stored, its value is taken as the weight of the undirected edge {i, j}. If both are stored, the value in the upper diagonal is taken. Off-diagonal entries which are not explicitly stored are treated as infinite, indicating absent edges. y : None There is no need for a target in a transformer, yet the pipeline API requires this parameter. Returns ------- self : object """ validate_params( self.get_params(), self._hyperparameters, exclude=["n_jobs"]) if isinstance(self.weights, str) and self.weights != "DTM": raise ValueError(f"'{self.weights}' passed for `weights` but the " f"only allowed string is 'DTM'.") self.effective_weight_params_ = {"p": 1} if self.weights == "DTM": key = "DTM" self.effective_weight_params_.update({"n_neighbors": 3, "r": 2}) else: key = "general" if self.weight_params: self.effective_weight_params_.update(self.weight_params) validate_params(self.effective_weight_params_, _AVAILABLE_RIPS_WEIGHTS[key]) self._is_precomputed = self.metric == "precomputed" check_point_clouds(X, accept_sparse=True, force_all_finite=True, distance_matrices=self._is_precomputed) if self.infinity_values is None: self.infinity_values_ = self.max_edge_weight else: self.infinity_values_ = self.infinity_values self._homology_dimensions = sorted(self.homology_dimensions) self._max_homology_dimension = self._homology_dimensions[-1] return self
[docs] def transform(self, X, y=None): """For each point cloud or distance matrix in `X`, compute the relevant persistence diagram as an array of triples [b, d, q]. Each triple represents a persistent topological feature in dimension q (belonging to `homology_dimensions`) which is born at b and dies at d. Only triples in which b < d are meaningful. Triples in which b and d are equal ("diagonal elements") may be artificially introduced during the computation for padding purposes, since the number of non-trivial persistent topological features is typically not constant across samples. They carry no information and hence should be effectively ignored by any further computation. Parameters ---------- X : ndarray or list of length n_samples Input data representing a collection of point clouds if `metric` was not set to ``"precomputed"``, and of distance matrices or adjacency matrices of weighted undirected graphs otherwise. Can be either a 3D ndarray whose zeroth dimension has size ``n_samples``, or a list containing ``n_samples`` 2D ndarrays/sparse matrices. Point cloud arrays have shape ``(n_points, n_dimensions)``, and if `X` is a list these shapes can vary between point clouds. If `metric` was set to ``"precomputed"``, then: - All entries of `X` should not contain infinities or negative values (contrary to :class:`VietorisRipsPersistence`). - The diagonals of entries of `X` are ignored (after the vertex weights are computed, when `weights` is a callable). - If entries of `X` are dense, only their upper diagonal portions are considered. - If entries of `X` are sparse, they do not need to be upper diagonal or symmetric. If only one of entry (i, j) and (j, i) is stored, its value is taken as the weight of the undirected edge {i, j}. If both are stored, the value in the upper diagonal is taken. Off-diagonal entries which are not explicitly stored are treated as infinite, indicating absent edges. y : None There is no need for a target in a transformer, yet the pipeline API requires this parameter. Returns ------- Xt : ndarray of shape (n_samples, n_features, 3) Array of persistence diagrams computed from the feature arrays or distance matrices in `X`. ``n_features`` equals :math:`\\sum_q n_q`, where :math:`n_q` is the maximum number of topological features in dimension :math:`q` across all samples in `X`. """ check_is_fitted(self) X = check_point_clouds(X, accept_sparse=True, force_all_finite=True, distance_matrices=self._is_precomputed) Xt = Parallel(n_jobs=self.n_jobs)( delayed(self._ripser_diagram)(x) for x in X) Xt = _postprocess_diagrams( Xt, "ripser", self._homology_dimensions, self.infinity_values_, self.reduced_homology ) return Xt
[docs] @staticmethod def plot(Xt, sample=0, homology_dimensions=None, plotly_params=None): """Plot a sample from a collection of persistence diagrams, with homology in multiple dimensions. Parameters ---------- Xt : ndarray of shape (n_samples, n_features, 3) Collection of persistence diagrams, such as returned by :meth:`transform`. sample : int, optional, default: ``0`` Index of the sample in `Xt` to be plotted. homology_dimensions : list, tuple or None, optional, default: ``None`` Which homology dimensions to include in the plot. ``None`` means plotting all dimensions present in ``Xt[sample]``. plotly_params : dict or None, optional, default: ``None`` Custom parameters to configure the plotly figure. Allowed keys are ``"traces"`` and ``"layout"``, and the corresponding values should be dictionaries containing keyword arguments as would be fed to the :meth:`update_traces` and :meth:`update_layout` methods of :class:`plotly.graph_objects.Figure`. Returns ------- fig : :class:`plotly.graph_objects.Figure` object Plotly figure. """ return plot_diagram( Xt[sample], homology_dimensions=homology_dimensions, plotly_params=plotly_params )
[docs]@adapt_fit_transform_docs class SparseRipsPersistence(BaseEstimator, TransformerMixin, PlotterMixin): """:ref:`Persistence diagrams <persistence_diagram>` resulting from :ref:`Sparse Vietoris–Rips filtrations <vietoris-rips_complex_and_vietoris-rips_persistence>`. Given a :ref:`point cloud <distance_matrices_and_point_clouds>` in Euclidean space, or an abstract :ref:`metric space <distance_matrices_and_point_clouds>` encoded by a distance matrix, information about the appearance and disappearance of topological features (technically, :ref:`homology classes <homology_and_cohomology>`) of various dimensions and at different scales is summarised in the corresponding persistence diagram. **Important note**: - Persistence diagrams produced by this class must be interpreted with care due to the presence of padding triples which carry no information. See :meth:`transform` for additional information. Parameters ---------- metric : string or callable, optional, default: ``"euclidean"`` If set to ``"precomputed"``, input data is to be interpreted as a collection of distance matrices. Otherwise, input data is to be interpreted as a collection of point clouds (i.e. feature arrays), and `metric` determines a rule with which to calculate distances between pairs of instances (i.e. rows) in these arrays. If `metric` is a string, it must be one of the options allowed by :func:`scipy.spatial.distance.pdist` for its metric parameter, or a metric listed in :obj:`sklearn.pairwise.PAIRWISE_DISTANCE_FUNCTIONS`, including "euclidean", "manhattan", or "cosine". If `metric` is a callable, it is called on each pair of instances and the resulting value recorded. The callable should take two arrays from the entry in `X` as input, and return a value indicating the distance between them. homology_dimensions : list or tuple, optional, default: ``(0, 1)`` Dimensions (non-negative integers) of the topological features to be detected. coeff : int prime, optional, default: ``2`` Compute homology with coefficients in the prime field :math:`\\mathbb{F}_p = \\{ 0, \\ldots, p - 1 \\}` where :math:`p` equals `coeff`. epsilon : float between 0. and 1., optional, default: ``0.1`` Parameter controlling the approximation to the exact Vietoris–Rips filtration. If set to `0.`, :class:`SparseRipsPersistence` leads to the same results as :class:`VietorisRipsPersistence` but is slower. max_edge_length : float, optional, default: ``numpy.inf`` Maximum value of the Sparse Rips filtration parameter. Points whose distance is greater than this value will never be connected by an edge, and topological features at scales larger than this value will not be detected. infinity_values : float or None, default: ``None`` Which death value to assign to features which are still alive at filtration value `max_edge_length`. ``None`` means that this death value is declared to be equal to `max_edge_length`. reduced_homology : bool, optional, default: ``True`` If ``True``, the earliest-born triple in homology dimension 0 which has infinite death is discarded from each diagram computed in :meth:`transform`. n_jobs : int or None, optional, default: ``None`` The number of jobs to use for the computation. ``None`` means 1 unless in a :obj:`joblib.parallel_backend` context. ``-1`` means using all processors. Attributes ---------- infinity_values_ : float Effective death value to assign to features which are still alive at filtration value `max_edge_length`. Set in :meth:`fit`. See also -------- VietorisRipsPersistence, WeightedRipsPersistence, FlagserPersistence, WeakAlphaPersistence, EuclideanCechPersistence, ConsistentRescaling, ConsecutiveRescaling Notes ----- `GUDHI <https://github.com/GUDHI/gudhi-devel>`_ is used as a C++ backend for computing sparse Vietoris–Rips persistent homology [1]_. Python bindings were modified for performance. References ---------- .. [1] C. Maria, "Persistent Cohomology", 2020; `GUDHI User and Reference Manual <http://gudhi.gforge.inria.fr/doc/3.1.0/group__persistent__\ cohomology.html>`_. """ _hyperparameters = { "metric": {"type": (str, FunctionType)}, "homology_dimensions": { "type": (list, tuple), "of": {"type": int, "in": Interval(0, np.inf, closed="left")} }, "coeff": {"type": int, "in": Interval(2, np.inf, closed="left")}, "epsilon": {"type": Real, "in": Interval(0, 1, closed="both")}, "max_edge_length": {"type": Real}, "infinity_values": {"type": (Real, type(None))}, "reduced_homology": {"type": bool} }
[docs] def __init__(self, metric="euclidean", homology_dimensions=(0, 1), coeff=2, epsilon=0.1, max_edge_length=np.inf, infinity_values=None, reduced_homology=True, n_jobs=None): self.metric = metric self.homology_dimensions = homology_dimensions self.coeff = coeff self.epsilon = epsilon self.max_edge_length = max_edge_length self.infinity_values = infinity_values self.reduced_homology = reduced_homology self.n_jobs = n_jobs
def _gudhi_diagram(self, X): Xdgm = pairwise_distances(X, metric=self.metric) sparse_rips_complex = SparseRipsComplex( distance_matrix=Xdgm, max_edge_length=self.max_edge_length, sparse=self.epsilon ) simplex_tree = sparse_rips_complex.create_simplex_tree( max_dimension=max(self._homology_dimensions) + 1 ) Xdgm = simplex_tree.persistence( homology_coeff_field=self.coeff, min_persistence=0 ) return Xdgm
[docs] def fit(self, X, y=None): """Calculate :attr:`infinity_values_`. Then, return the estimator. This method is here to implement the usual scikit-learn API and hence work in pipelines. Parameters ---------- X : ndarray or list of length n_samples Input data representing a collection of point clouds if `metric` was not set to ``"precomputed"``, and of distance matrices otherwise. Can be either a 3D ndarray whose zeroth dimension has size ``n_samples``, or a list containing ``n_samples`` 2D ndarrays. Point cloud arrays have shape ``(n_points, n_dimensions)``, and if `X` is a list these shapes can vary between point clouds. If `metric` was set to ``"precomputed"``, each entry of `X` should be compatible with a filtration, i.e. the value at index (i, j) should be no smaller than the values at diagonal indices (i, i) and (j, j). y : None There is no need for a target in a transformer, yet the pipeline API requires this parameter. Returns ------- self : object """ validate_params( self.get_params(), self._hyperparameters, exclude=["n_jobs"]) self._is_precomputed = self.metric == "precomputed" check_point_clouds(X, accept_sparse=True, distance_matrices=self._is_precomputed) if self.infinity_values is None: self.infinity_values_ = self.max_edge_length else: self.infinity_values_ = self.infinity_values self._homology_dimensions = sorted(self.homology_dimensions) self._max_homology_dimension = self._homology_dimensions[-1] return self
[docs] def transform(self, X, y=None): """For each point cloud or distance matrix in `X`, compute the relevant persistence diagram as an array of triples [b, d, q]. Each triple represents a persistent topological feature in dimension q (belonging to `homology_dimensions`) which is born at b and dies at d. Only triples in which b < d are meaningful. Triples in which b and d are equal ("diagonal elements") may be artificially introduced during the computation for padding purposes, since the number of non-trivial persistent topological features is typically not constant across samples. They carry no information and hence should be effectively ignored by any further computation. Parameters ---------- X : ndarray or list of length n_samples Input data representing a collection of point clouds if `metric` was not set to ``"precomputed"``, and of distance matrices otherwise. Can be either a 3D ndarray whose zeroth dimension has size ``n_samples``, or a list containing ``n_samples`` 2D ndarrays. Point cloud arrays have shape ``(n_points, n_dimensions)``, and if `X` is a list these shapes can vary between point clouds. If `metric` was set to ``"precomputed"``, each entry of `X` should be compatible with a filtration, i.e. the value at index (i, j) should be no smaller than the values at diagonal indices (i, i) and (j, j). y : None There is no need for a target in a transformer, yet the pipeline API requires this parameter. Returns ------- Xt : ndarray of shape (n_samples, n_features, 3) Array of persistence diagrams computed from the feature arrays or distance matrices in `X`. ``n_features`` equals :math:`\\sum_q n_q`, where :math:`n_q` is the maximum number of topological features in dimension :math:`q` across all samples in `X`. """ check_is_fitted(self) X = check_point_clouds(X, accept_sparse=True, distance_matrices=self._is_precomputed) Xt = Parallel(n_jobs=self.n_jobs)( delayed(self._gudhi_diagram)(x) for x in X) Xt = _postprocess_diagrams( Xt, "gudhi", self._homology_dimensions, self.infinity_values_, self.reduced_homology ) return Xt
[docs] @staticmethod def plot(Xt, sample=0, homology_dimensions=None, plotly_params=None): """Plot a sample from a collection of persistence diagrams, with homology in multiple dimensions. Parameters ---------- Xt : ndarray of shape (n_samples, n_features, 3) Collection of persistence diagrams, such as returned by :meth:`transform`. sample : int, optional, default: ``0`` Index of the sample in `Xt` to be plotted. homology_dimensions : list, tuple or None, optional, default: ``None`` Which homology dimensions to include in the plot. ``None`` means plotting all dimensions present in ``Xt[sample]``. plotly_params : dict or None, optional, default: ``None`` Custom parameters to configure the plotly figure. Allowed keys are ``"traces"`` and ``"layout"``, and the corresponding values should be dictionaries containing keyword arguments as would be fed to the :meth:`update_traces` and :meth:`update_layout` methods of :class:`plotly.graph_objects.Figure`. Returns ------- fig : :class:`plotly.graph_objects.Figure` object Plotly figure. """ return plot_diagram( Xt[sample], homology_dimensions=homology_dimensions, plotly_params=plotly_params )
[docs]@adapt_fit_transform_docs class WeakAlphaPersistence(BaseEstimator, TransformerMixin, PlotterMixin): """:ref:`Persistence diagrams <persistence_diagram>` resulting from :ref:`weak alpha filtrations <TODO>`. Given a :ref:`point cloud <distance_matrices_and_point_clouds>` in Euclidean space, information about the appearance and disappearance of topological features (technically, :ref:`homology classes <homology_and_cohomology>`) of various dimensions and at different scales is summarised in the corresponding persistence diagram. The weak alpha filtration of a point cloud is defined to be the :ref:`Vietoris–Rips filtration <vietoris-rips_complex_and_vietoris-rips_persistence>` of the sparse matrix of Euclidean distances between neighbouring vertices in the Delaunay triangulation of the point cloud. In low dimensions, computing the persistent homology of this filtration can be much faster than computing Vietoris–Rips persistent homology via :class:`VietorisRipsPersistence`. **Important note**: - Persistence diagrams produced by this class must be interpreted with care due to the presence of padding triples which carry no information. See :meth:`transform` for additional information. Parameters ---------- homology_dimensions : list or tuple, optional, default: ``(0, 1)`` Dimensions (non-negative integers) of the topological features to be detected. coeff : int prime, optional, default: ``2`` Compute homology with coefficients in the prime field :math:`\\mathbb{F}_p = \\{ 0, \\ldots, p - 1 \\}` where :math:`p` equals `coeff`. max_edge_length : float, optional, default: ``numpy.inf`` Maximum value of the Vietoris–Rips filtration parameter. Points whose distance is greater than this value will never be connected by an edge, and topological features at scales larger than this value will not be detected. infinity_values : float or None, default: ``None`` Which death value to assign to features which are still alive at filtration value `max_edge_length`. ``None`` means that this death value is declared to be equal to `max_edge_length`. reduced_homology : bool, optional, default: ``True`` If ``True``, the earliest-born triple in homology dimension 0 which has infinite death is discarded from each diagram computed in :meth:`transform`. n_jobs : int or None, optional, default: ``None`` The number of jobs to use for the computation. ``None`` means 1 unless in a :obj:`joblib.parallel_backend` context. ``-1`` means using all processors. Attributes ---------- infinity_values_ : float Effective death value to assign to features which are still alive at filtration value `max_edge_length`. See also -------- VietorisRipsPersistence, WeightedRipsPersistence, SparseRipsPersistence, FlagserPersistence, EuclideanCechPersistence Notes ----- Delaunay triangulation are computed by :class:`scipy.spatial.Delaunay`. `Ripser <https://github.com/Ripser/ripser>`_ [1]_ is used as a C++ backend for computing Vietoris–Rips persistent homology. Python bindings were modified for performance from the `ripser.py <https://github.com/scikit-tda/ripser.py>`_ package. References ---------- .. [1] U. Bauer, "Ripser: efficient computation of Vietoris–Rips persistence barcodes", 2019; `arXiv:1908.02518 <https://arxiv.org/abs/1908.02518>`_. """ _hyperparameters = { "homology_dimensions": { "type": (list, tuple), "of": {"type": int, "in": Interval(0, np.inf, closed="left")} }, "coeff": {"type": int, "in": Interval(2, np.inf, closed="left")}, "max_edge_length": {"type": Real}, "infinity_values": {"type": (Real, type(None))}, "reduced_homology": {"type": bool} }
[docs] def __init__(self, homology_dimensions=(0, 1), coeff=2, max_edge_length=np.inf, infinity_values=None, reduced_homology=True, n_jobs=None): self.homology_dimensions = homology_dimensions self.coeff = coeff self.max_edge_length = max_edge_length self.infinity_values = infinity_values self.reduced_homology = reduced_homology self.n_jobs = n_jobs
def _weak_alpha_diagram(self, X): # `indices` will serve as the array of column indices indptr, indices = Delaunay(X).vertex_neighbor_vertices # Compute the array of row indices row = np.zeros_like(indices) row[indptr[1:-1]] = 1 np.cumsum(row, out=row) # We only need the upper diagonal mask = indices > row row, col = row[mask], indices[mask] dists = np.linalg.norm(X[row] - X[col], axis=1) # Note: passing the shape explicitly should not be needed in more # recent versions of C++ ripser n_points = len(X) dm = coo_matrix((dists, (row, col)), shape=(n_points, n_points)) Xdgms = ripser(dm, maxdim=self._max_homology_dimension, thresh=self.max_edge_length, coeff=self.coeff, metric="precomputed")["dgms"] return Xdgms
[docs] def fit(self, X, y=None): """Calculate :attr:`infinity_values_`. Then, return the estimator. This method is here to implement the usual scikit-learn API and hence work in pipelines. Parameters ---------- X : ndarray or list of length n_samples Input data representing a collection of point clouds. Can be either a 3D ndarray whose zeroth dimension has size ``n_samples``, or a list containing ``n_samples`` 2D ndarrays. Point cloud arrays have shape ``(n_points, n_dimensions)``, and if `X` is a list these shapes can vary between point clouds. y : None There is no need for a target in a transformer, yet the pipeline API requires this parameter. Returns ------- self : object """ validate_params( self.get_params(), self._hyperparameters, exclude=["n_jobs"]) check_point_clouds(X) if self.infinity_values is None: self.infinity_values_ = self.max_edge_length else: self.infinity_values_ = self.infinity_values self._homology_dimensions = sorted(self.homology_dimensions) self._max_homology_dimension = self._homology_dimensions[-1] return self
[docs] def transform(self, X, y=None): """For each point cloud in `X`, compute the relevant persistence diagram as an array of triples [b, d, q]. Each triple represents a persistent topological feature in dimension q (belonging to `homology_dimensions`) which is born at b and dies at d. Only triples in which b < d are meaningful. Triples in which b and d are equal ("diagonal elements") may be artificially introduced during the computation for padding purposes, since the number of non-trivial persistent topological features is typically not constant across samples. They carry no information and hence should be effectively ignored by any further computation. Parameters ---------- X : ndarray or list of length n_samples Input data representing a collection of point clouds. Can be either a 3D ndarray whose zeroth dimension has size ``n_samples``, or a list containing ``n_samples`` 2D ndarrays. Point cloud arrays have shape ``(n_points, n_dimensions)``, and if `X` is a list these shapes can vary between point clouds. y : None There is no need for a target in a transformer, yet the pipeline API requires this parameter. Returns ------- Xt : ndarray of shape (n_samples, n_features, 3) Array of persistence diagrams computed from the feature arrays or distance matrices in `X`. ``n_features`` equals :math:`\\sum_q n_q`, where :math:`n_q` is the maximum number of topological features in dimension :math:`q` across all samples in `X`. """ check_is_fitted(self) X = check_point_clouds(X) Xt = Parallel(n_jobs=self.n_jobs)( delayed(self._weak_alpha_diagram)(x) for x in X) Xt = _postprocess_diagrams( Xt, "ripser", self._homology_dimensions, self.infinity_values_, self.reduced_homology ) return Xt
[docs] @staticmethod def plot(Xt, sample=0, homology_dimensions=None, plotly_params=None): """Plot a sample from a collection of persistence diagrams, with homology in multiple dimensions. Parameters ---------- Xt : ndarray of shape (n_samples, n_features, 3) Collection of persistence diagrams, such as returned by :meth:`transform`. sample : int, optional, default: ``0`` Index of the sample in `Xt` to be plotted. homology_dimensions : list, tuple or None, optional, default: ``None`` Which homology dimensions to include in the plot. ``None`` means plotting all dimensions present in ``Xt[sample]``. plotly_params : dict or None, optional, default: ``None`` Custom parameters to configure the plotly figure. Allowed keys are ``"traces"`` and ``"layout"``, and the corresponding values should be dictionaries containing keyword arguments as would be fed to the :meth:`update_traces` and :meth:`update_layout` methods of :class:`plotly.graph_objects.Figure`. Returns ------- fig : :class:`plotly.graph_objects.Figure` object Plotly figure. """ return plot_diagram( Xt[sample], homology_dimensions=homology_dimensions, plotly_params=plotly_params )
[docs]@adapt_fit_transform_docs class EuclideanCechPersistence(BaseEstimator, TransformerMixin, PlotterMixin): """:ref:`Persistence diagrams <persistence_diagram>` resulting from `Cech filtrations <cech_complex_and_cech_persistence>`_. Given a :ref:`point cloud <distance_matrices_and_point_clouds>` in Euclidean space, information about the appearance and disappearance of topological features (technically, :ref:`homology classes <homology_and_cohomology>`) of various dimensions and at different scales is summarised in the corresponding persistence diagram. **Important note**: - Persistence diagrams produced by this class must be interpreted with care due to the presence of padding triples which carry no information. See :meth:`transform` for additional information. Parameters ---------- homology_dimensions : list or tuple, optional, default: ``(0, 1)`` Dimensions (non-negative integers) of the topological features to be detected. coeff : int prime, optional, default: ``2`` Compute homology with coefficients in the prime field :math:`\\mathbb{F}_p = \\{ 0, \\ldots, p - 1 \\}` where :math:`p` equals `coeff`. max_edge_length : float, optional, default: ``numpy.inf`` Maximum value of the Cech filtration parameter. Topological features at scales larger than this value will not be detected. infinity_values : float or None, default: ``None`` Which death value to assign to features which are still alive at filtration value `max_edge_length`. ``None`` means that this death value is declared to be equal to `max_edge_length`. reduced_homology : bool, optional, default: ``True`` If ``True``, the earliest-born triple in homology dimension 0 which has infinite death is discarded in :meth:`transform`. n_jobs : int or None, optional, default: ``None`` The number of jobs to use for the computation. ``None`` means 1 unless in a :obj:`joblib.parallel_backend` context. ``-1`` means using all processors. Attributes ---------- infinity_values_ : float Effective death value to assign to features which are still alive at filtration value `max_edge_length`. See also -------- VietorisRipsPersistence, FlagserPersistence, SparseRipsPersistence, WeakAlphaPersistence Notes ----- `GUDHI <https://github.com/GUDHI/gudhi-devel>`_ is used as a C++ backend for computing Cech persistent homology [1]_. Python bindings were modified for performance. References ---------- .. [1] C. Maria, "Persistent Cohomology", 2020; `GUDHI User and Reference Manual <http://gudhi.gforge.inria.fr/doc/3.1.0/group__persistent__\ cohomology.html>`_. """ _hyperparameters = { "homology_dimensions": { "type": (list, tuple), "of": {"type": int, "in": Interval(0, np.inf, closed="left")} }, "coeff": {"type": int, "in": Interval(2, np.inf, closed="left")}, "max_edge_length": {"type": Real, "in": Interval(0, np.inf, closed="right")}, "infinity_values": {"type": (Real, type(None)), "in": Interval(0, np.inf, closed="neither")}, "reduced_homology": {"type": bool} }
[docs] def __init__(self, homology_dimensions=(0, 1), coeff=2, max_edge_length=np.inf, infinity_values=None, reduced_homology=True, n_jobs=None): self.homology_dimensions = homology_dimensions self.coeff = coeff self.max_edge_length = max_edge_length self.infinity_values = infinity_values self.reduced_homology = reduced_homology self.n_jobs = n_jobs
def _gudhi_diagram(self, X): cech_complex = CechComplex(points=X, max_radius=self.max_edge_length) simplex_tree = cech_complex.create_simplex_tree( max_dimension=max(self._homology_dimensions) + 1 ) Xdgm = simplex_tree.persistence(homology_coeff_field=self.coeff, min_persistence=0) return Xdgm
[docs] def fit(self, X, y=None): """Calculate :attr:`infinity_values_`. Then, return the estimator. This method is here to implement the usual scikit-learn API and hence work in pipelines. Parameters ---------- X : ndarray or list of length n_samples Input data representing a collection of point clouds. Can be either a 3D ndarray whose zeroth dimension has size ``n_samples``, or a list containing ``n_samples`` 2D ndarrays. Point cloud arrays have shape ``(n_points, n_dimensions)``, and if `X` is a list these shapes can vary between point clouds. y : None There is no need for a target in a transformer, yet the pipeline API requires this parameter. Returns ------- self : object """ check_point_clouds(X) validate_params( self.get_params(), self._hyperparameters, exclude=["n_jobs"]) if self.infinity_values is None: self.infinity_values_ = self.max_edge_length else: self.infinity_values_ = self.infinity_values self._homology_dimensions = sorted(self.homology_dimensions) self._max_homology_dimension = self._homology_dimensions[-1] return self
[docs] def transform(self, X, y=None): """For each point cloud in `X`, compute the relevant persistence diagram as an array of triples [b, d, q]. Each triple represents a persistent topological feature in dimension q (belonging to `homology_dimensions`) which is born at b and dies at d. Only triples in which b < d are meaningful. Triples in which b and d are equal ("diagonal elements") may be artificially introduced during the computation for padding purposes, since the number of non-trivial persistent topological features is typically not constant across samples. They carry no information and hence should be effectively ignored by any further computation. Parameters ---------- X : ndarray or list of length n_samples Input data representing a collection of point clouds. Can be either a 3D ndarray whose zeroth dimension has size ``n_samples``, or a list containing ``n_samples`` 2D ndarrays. Point cloud arrays have shape ``(n_points, n_dimensions)``, and if `X` is a list these shapes can vary between point clouds. y : None There is no need for a target in a transformer, yet the pipeline API requires this parameter. Returns ------- Xt : ndarray of shape (n_samples, n_features, 3) Array of persistence diagrams computed from the feature arrays in `X`. ``n_features`` equals :math:`\\sum_q n_q`, where :math:`n_q` is the maximum number of topological features in dimension :math:`q` across all samples in `X`. """ check_is_fitted(self) X = check_point_clouds(X) Xt = Parallel(n_jobs=self.n_jobs)(delayed(self._gudhi_diagram)(x) for x in X) Xt = _postprocess_diagrams( Xt, "gudhi", self._homology_dimensions, self.infinity_values_, self.reduced_homology ) return Xt
[docs] @staticmethod def plot(Xt, sample=0, homology_dimensions=None, plotly_params=None): """Plot a sample from a collection of persistence diagrams, with homology in multiple dimensions. Parameters ---------- Xt : ndarray of shape (n_samples, n_features, 3) Collection of persistence diagrams, such as returned by :meth:`transform`. sample : int, optional, default: ``0`` Index of the sample in `Xt` to be plotted. homology_dimensions : list, tuple or None, optional, default: ``None`` Which homology dimensions to include in the plot. ``None`` means plotting all dimensions present in ``Xt[sample]``. plotly_params : dict or None, optional, default: ``None`` Custom parameters to configure the plotly figure. Allowed keys are ``"traces"`` and ``"layout"``, and the corresponding values should be dictionaries containing keyword arguments as would be fed to the :meth:`update_traces` and :meth:`update_layout` methods of :class:`plotly.graph_objects.Figure`. Returns ------- fig : :class:`plotly.graph_objects.Figure` object Plotly figure. """ return plot_diagram( Xt[sample], homology_dimensions=homology_dimensions, plotly_params=plotly_params )
[docs]@adapt_fit_transform_docs class FlagserPersistence(BaseEstimator, TransformerMixin, PlotterMixin): """:ref:`Persistence diagrams <persistence_diagram>` resulting from :ref:`filtrations <filtered_complex>` of :ref:`directed or undirected flag complexes <clique_and_flag_complexes>` [1]_. Given a weighted directed or undirected graph, information about the appearance and disappearance of topological features (technically, :ref:`homology classes <homology_and_cohomology>`) of various dimension and at different scales is summarised in the corresponding persistence diagram. **Important note**: - Persistence diagrams produced by this class must be interpreted with care due to the presence of padding triples which carry no information. See :meth:`transform` for additional information. Parameters ---------- homology_dimensions : list or tuple, optional, default: ``(0, 1)`` Dimensions (non-negative integers) of the topological features to be detected. directed : bool, optional, default: ``True`` If ``True``, :meth:`transform` computes the persistence diagrams of the filtered directed flag complexes arising from the input collection of weighted directed graphs. If ``False``, :meth:`transform` computes the persistence diagrams of the filtered undirected flag complexes obtained by regarding all input weighted graphs as undirected, and: - if `max_edge_weight` is ``numpy.inf``, it is sufficient to pass a collection of (dense or sparse) upper-triangular matrices; - if `max_edge_weight` is finite, it is recommended to pass either a collection of symmetric dense matrices, or a collection of sparse upper-triangular matrices. filtration : string, optional, default: ``"max"`` Algorithm determining the filtration values of higher order simplices from the weights of the vertices and edges. Possible values are: ["dimension", "zero", "max", "max3", "max_plus_one", "product", "sum", "pmean", "pmoment", "remove_edges", "vertex_degree"] coeff : int prime, optional, default: ``2`` Compute homology with coefficients in the prime field :math:`\\mathbb{F}_p = \\{ 0, \\ldots, p - 1 \\}` where :math:`p` equals `coeff`. max_edge_weight : float, optional, default: ``numpy.inf`` Maximum edge weight to be considered in the filtration. All edge weights greater than this value will be considered as absent from the filtration and topological features at scales larger than this value will not be detected. infinity_values : float or None, default: ``None`` Which death value to assign to features which are still alive at filtration value `max_edge_weight`. ``None`` means that this death value is declared to be equal to `max_edge_weight`. reduced_homology : bool, optional, default: ``True`` If ``True``, the earliest-born triple in homology dimension 0 which has infinite death is discarded from each diagram computed in :meth:`transform`. max_entries : int, optional, default: ``-1`` Number controlling the degree of precision in the matrix reductions performed by the the backend. Corresponds to the parameter ``approximation`` in :func:`pyflagser.flagser_weighted` and :func:`pyflagser.flagser_unweighted`. Increase for higher precision, decrease for faster computation. A good value is often ``100000`` in hard problems. A negative value computes highest possible precision. n_jobs : int or None, optional, default: ``None`` The number of jobs to use for the computation. ``None`` means 1 unless in a :obj:`joblib.parallel_backend` context. ``-1`` means using all processors. Attributes ---------- infinity_values_ : float Effective death value to assign to features which are still alive at filtration value `max_edge_weight`. See also -------- VietorisRipsPersistence, WeightedRipsPersistence, SparseRipsPersistence, WeakAlphaPersistence, EuclideanCechPersistence, ConsistentRescaling, ConsecutiveRescaling Notes ----- The `pyflagser <https://github.com/giotto-ai/pyflagser>`_ Python package is used for binding `Flagser <https://github.com/luetge/flagser>`_, a C++ backend for computing the (persistent) homology of (filtered) directed flag complexes. For more details, please refer to the `flagser \ documentation <https://github.com/luetge/flagser/blob/master/docs/\ documentation_flagser.pdf>`_. References ---------- .. [1] D. Luetgehetmann, D. Govc, J. P. Smith, and R. Levi, "Computing persistent homology of directed flag complexes", *Algorithms*, 13(1), 2020. """ _hyperparameters = { "homology_dimensions": { "type": (list, tuple), "of": {"type": int, "in": Interval(0, np.inf, closed="left")} }, "directed": {"type": bool}, "coeff": {"type": int, "in": Interval(2, np.inf, closed="left")}, "max_edge_weight": {"type": Real}, "infinity_values": {"type": (Real, type(None))}, "reduced_homology": {"type": bool}, "max_entries": {"type": int} }
[docs] def __init__(self, homology_dimensions=(0, 1), directed=True, filtration="max", coeff=2, max_edge_weight=np.inf, infinity_values=None, reduced_homology=True, max_entries=-1, n_jobs=None): self.homology_dimensions = homology_dimensions self.directed = directed self.filtration = filtration self.coeff = coeff self.max_edge_weight = max_edge_weight self.infinity_values = infinity_values self.reduced_homology = reduced_homology self.max_entries = max_entries self.n_jobs = n_jobs
def _flagser_diagram(self, X): Xdgms = [np.empty((0, 2), dtype=float)] * self._min_homology_dimension Xdgms += flagser_weighted(X, max_edge_weight=self.max_edge_weight, min_dimension=self._min_homology_dimension, max_dimension=self._max_homology_dimension, directed=self.directed, filtration=self.filtration, coeff=self.coeff, approximation=self.max_entries)["dgms"] n_missing_dims = self._max_homology_dimension + 1 - len(Xdgms) if n_missing_dims: Xdgms += [np.empty((0, 2), dtype=float)] * n_missing_dims return Xdgms
[docs] def fit(self, X, y=None): """Calculate :attr:`infinity_values_`. Then, return the estimator. This method is here to implement the usual scikit-learn API and hence work in pipelines. Parameters ---------- X : ndarray or list of length n_samples Input collection of adjacency matrices of weighted directed or undirected graphs. Can be either a 3D ndarray whose zeroth dimension has size ``n_samples``, or a list containing ``n_samples`` 2D ndarrays/sparse matrices. In each adjacency matrix, diagonal elements are vertex weights and off-diagonal elements are edge weights. It is assumed that a vertex weight cannot be larger than the weight of the edges it forms. The way zero values are handled depends on the format of the matrix. If the matrix is a dense ``numpy.ndarray``, zero values denote zero-weighted edges. If the matrix is a sparse ``scipy.sparse`` matrix, explicitly stored off-diagonal zeros and all diagonal zeros denote zero-weighted edges. Off-diagonal values that have not been explicitly stored are treated by ``scipy.sparse`` as zeros but will be understood as infinitely-valued edges, i.e., edges absent from the filtration. y : None There is no need for a target in a transformer, yet the pipeline API requires this parameter. Returns ------- self : object """ check_point_clouds(X, accept_sparse=True, distance_matrices=True) validate_params( self.get_params(), self._hyperparameters, exclude=["n_jobs", "filtration"]) if self.infinity_values is None: self.infinity_values_ = self.max_edge_weight else: self.infinity_values_ = self.infinity_values self._homology_dimensions = sorted(self.homology_dimensions) self._min_homology_dimension = self._homology_dimensions[0] self._max_homology_dimension = self._homology_dimensions[-1] return self
[docs] def transform(self, X, y=None): """For each adjacency matrix in `X`, compute the relevant persistence diagram as an array of triples [b, d, q]. Each triple represents a persistent topological feature in dimension q (belonging to `homology_dimensions`) which is born at b and dies at d. Only triples in which b < d are meaningful. Triples in which b and d are equal ("diagonal elements") may be artificially introduced during the computation for padding purposes, since the number of non-trivial persistent topological features is typically not constant across samples. They carry no information and hence should be effectively ignored by any further computation. Parameters ---------- X : ndarray or list of length n_samples Input collection of adjacency matrices of weighted directed or undirected graphs. Can be either a 3D ndarray whose zeroth dimension has size ``n_samples``, or a list containing ``n_samples`` 2D ndarrays/sparse matrices. In each adjacency matrix, diagonal elements are vertex weights and off-diagonal elements are edges weights. It is assumed that a vertex weight cannot be larger than the weight of the edges it forms. The way zero values are handled depends on the format of the matrix. If the matrix is a dense ``numpy.ndarray``, zero values denote zero-weighted edges. If the matrix is a sparse ``scipy.sparse`` matrix, explicitly stored off-diagonal zeros and all diagonal zeros denote zero-weighted edges. Off-diagonal values that have not been explicitly stored are treated by ``scipy.sparse`` as zeros but will be understood as infinitely-valued edges, i.e., edges absent from the filtration. y : None There is no need for a target in a transformer, yet the pipeline API requires this parameter. Returns ------- Xt : ndarray of shape (n_samples, n_features, 3) Array of persistence diagrams computed from the feature arrays or distance matrices in `X`. ``n_features`` equals :math:`\\sum_q n_q`, where :math:`n_q` is the maximum number of topological features in dimension :math:`q` across all samples in `X`. """ check_is_fitted(self) X = check_point_clouds(X, accept_sparse=True, distance_matrices=True) Xt = Parallel(n_jobs=self.n_jobs)( delayed(self._flagser_diagram)(x) for x in X) Xt = _postprocess_diagrams( Xt, "flagser", self._homology_dimensions, self.infinity_values_, self.reduced_homology ) return Xt
[docs] @staticmethod def plot(Xt, sample=0, homology_dimensions=None, plotly_params=None): """Plot a sample from a collection of persistence diagrams, with homology in multiple dimensions. Parameters ---------- Xt : ndarray of shape (n_samples, n_features, 3) Collection of persistence diagrams, such as returned by :meth:`transform`. sample : int, optional, default: ``0`` Index of the sample in `Xt` to be plotted. homology_dimensions : list, tuple or None, optional, default: ``None`` Which homology dimensions to include in the plot. ``None`` means plotting all dimensions present in ``Xt[sample]``. plotly_params : dict or None, optional, default: ``None`` Custom parameters to configure the plotly figure. Allowed keys are ``"traces"`` and ``"layout"``, and the corresponding values should be dictionaries containing keyword arguments as would be fed to the :meth:`update_traces` and :meth:`update_layout` methods of :class:`plotly.graph_objects.Figure`. Returns ------- fig : :class:`plotly.graph_objects.Figure` object Plotly figure. """ return plot_diagram( Xt[sample], homology_dimensions=homology_dimensions, plotly_params=plotly_params )