stats

Bases: dipy.segment.metricspeed.SumPointwiseEuclideanMetric

Computes the average of pointwise Euclidean distances between two sequential data.

A sequence of N-dimensional points is represented as a 2D array with shape (nb_points, nb_dimensions). A feature object can be specified in order to calculate the distance between the features, rather than directly between the sequential data.

Parameters

featureFeature object, optional

It is used to extract features before computing the distance.

Notes

The distance between two 2D sequential data:

s1       s2

0*   a    *0
  \       |
   \      |
   1*     |
    |  b  *1
    |      \
    2*      \
        c    *2

is equal to \((a+b+c)/3\) where \(a\) is the Euclidean distance between s1[0] and s2[0], \(b\) between s1[1] and s2[1] and \(c\) between s1[2] and s2[2].

Attributes

feature

Feature object used to extract features from sequential data

is_order_invariant

Is this metric invariant to the sequence’s ordering

Methods

__init__(*args, **kwargs)

Initialize self. See help(type(self)) for accurate signature.

Bases: dipy.segment.clustering.Clustering

Clusters streamlines using QuickBundles [Garyfallidis12].

Given a list of streamlines, the QuickBundles algorithm sequentially assigns each streamline to its closest bundle in \(\mathcal{O}(Nk)\) where \(N\) is the number of streamlines and \(k\) is the final number of bundles. If for a given streamline its closest bundle is farther than threshold, a new bundle is created and the streamline is assigned to it except if the number of bundles has already exceeded max_nb_clusters.

Parameters

thresholdfloat

The maximum distance from a bundle for a streamline to be still considered as part of it.

metricstr or Metric object (optional)

The distance metric to use when comparing two streamlines. By default, the Minimum average Direct-Flip (MDF) distance [Garyfallidis12] is used and streamlines are automatically resampled so they have 12 points.

max_nb_clustersint

Limits the creation of bundles.

References

Garyfallidis12(1,2,3)

Garyfallidis E. et al., QuickBundles a method for tractography simplification, Frontiers in Neuroscience, vol 6, no 175, 2012.

Examples

>>> from dipy.segment.clustering import QuickBundles
>>> from dipy.data import get_fnames
>>> from dipy.io.streamline import load_tractogram
>>> from dipy.tracking.streamline import Streamlines
>>> fname = get_fnames('fornix')
>>> fornix = load_tractogram(fname, 'same',
...                          bbox_valid_check=False).streamlines
>>> streamlines = Streamlines(fornix)
>>> # Segment fornix with a threshold of 10mm and streamlines resampled
>>> # to 12 points.
>>> qb = QuickBundles(threshold=10.)
>>> clusters = qb.cluster(streamlines)
>>> len(clusters)
4
>>> list(map(len, clusters))
[61, 191, 47, 1]
>>> # Resampling streamlines differently is done explicitly as follows.
>>> # Note this has an impact on the speed and the accuracy (tradeoff).
>>> from dipy.segment.metric import ResampleFeature
>>> from dipy.segment.metric import AveragePointwiseEuclideanMetric
>>> feature = ResampleFeature(nb_points=2)
>>> metric = AveragePointwiseEuclideanMetric(feature)
>>> qb = QuickBundles(threshold=10., metric=metric)
>>> clusters = qb.cluster(streamlines)
>>> len(clusters)
4
>>> list(map(len, clusters))
[58, 142, 72, 28]

Methods

__init__(threshold, metric='MDF_12points', max_nb_clusters=2147483647)

Initialize self. See help(type(self)) for accurate signature.

cluster(streamlines, ordering=None)

Clusters streamlines into bundles.

Performs quickbundles algorithm using predefined metric and threshold.

Parameters

streamlineslist of 2D arrays

Each 2D array represents a sequence of 3D points (points, 3).

orderingiterable of indices

Specifies the order in which data points will be clustered.

Returns

ClusterMapCentroid object

Result of the clustering.

Bases: object

kd-tree for quick nearest-neighbor lookup

This class provides an index into a set of k-dimensional points which can be used to rapidly look up the nearest neighbors of any point.

The algorithm used is described in Maneewongvatana and Mount 1999. The general idea is that the kd-tree is a binary trie, each of whose nodes represents an axis-aligned hyperrectangle. Each node specifies an axis and splits the set of points based on whether their coordinate along that axis is greater than or less than a particular value.

During construction, the axis and splitting point are chosen by the “sliding midpoint” rule, which ensures that the cells do not all become long and thin.

The tree can be queried for the r closest neighbors of any given point (optionally returning only those within some maximum distance of the point). It can also be queried, with a substantial gain in efficiency, for the r approximate closest neighbors.

For large dimensions (20 is already large) do not expect this to run significantly faster than brute force. High-dimensional nearest-neighbor queries are a substantial open problem in computer science.

Parameters

dataarray_like, shape (n,m)

The n data points of dimension m to be indexed. This array is not copied unless this is necessary to produce a contiguous array of doubles, and so modifying this data will result in bogus results. The data are also copied if the kd-tree is built with copy_data=True.

leafsizepositive int, optional

The number of points at which the algorithm switches over to brute-force. Default: 16.

compact_nodesbool, optional

If True, the kd-tree is built to shrink the hyperrectangles to the actual data range. This usually gives a more compact tree that is robust against degenerated input data and gives faster queries at the expense of longer build time. Default: True.

copy_databool, optional

If True the data is always copied to protect the kd-tree against data corruption. Default: False.

balanced_treebool, optional

If True, the median is used to split the hyperrectangles instead of the midpoint. This usually gives a more compact tree and faster queries at the expense of longer build time. Default: True.

boxsizearray_like or scalar, optional

Apply a m-d toroidal topology to the KDTree.. The topology is generated by \(x_i + n_i L_i\) where \(n_i\) are integers and \(L_i\) is the boxsize along i-th dimension. The input data shall be wrapped into \([0, L_i)\). A ValueError is raised if any of the data is outside of this bound.

Attributes

datandarray, shape (n,m)

The n data points of dimension m to be indexed. This array is not copied unless this is necessary to produce a contiguous array of doubles. The data are also copied if the kd-tree is built with copy_data=True.

leafsizepositive int

The number of points at which the algorithm switches over to brute-force.

mint

The dimension of a single data-point.

nint

The number of data points.

maxesndarray, shape (m,)

The maximum value in each dimension of the n data points.

minsndarray, shape (m,)

The minimum value in each dimension of the n data points.

treeobject, class cKDTreeNode

This attribute exposes a Python view of the root node in the cKDTree object. A full Python view of the kd-tree is created dynamically on the first access. This attribute allows you to create your own query functions in Python.

sizeint

The number of nodes in the tree.

Methods

__init__(*args, **kwargs)

Initialize self. See help(type(self)) for accurate signature.

boxsize

count_neighbors(self, other, r, p=2., weights=None, cumulative=True)

Count how many nearby pairs can be formed. (pair-counting)

Count the number of pairs (x1,x2) can be formed, with x1 drawn from self and x2 drawn from other, and where distance(x1, x2, p) <= r.

Data points on self and other are optionally weighted by the weights argument. (See below)

The algorithm we implement here is based on [1]. See notes for further discussion.

Parameters

othercKDTree instance

The other tree to draw points from, can be the same tree as self.

rfloat or one-dimensional array of floats

The radius to produce a count for. Multiple radii are searched with a single tree traversal. If the count is non-cumulative(cumulative=False), r defines the edges of the bins, and must be non-decreasing.

pfloat, optional

1<=p<=infinity. Which Minkowski p-norm to use. Default 2.0. A finite large p may cause a ValueError if overflow can occur.

weightstuple, array_like, or None, optional

If None, the pair-counting is unweighted. If given as a tuple, weights[0] is the weights of points in self, and weights[1] is the weights of points in other; either can be None to indicate the points are unweighted. If given as an array_like, weights is the weights of points in self and other. For this to make sense, self and other must be the same tree. If self and other are two different trees, a ValueError is raised. Default: None

cumulativebool, optional

Whether the returned counts are cumulative. When cumulative is set to False the algorithm is optimized to work with a large number of bins (>10) specified by r. When cumulative is set to True, the algorithm is optimized to work with a small number of r. Default: True

Returns

resultscalar or 1-D array

The number of pairs. For unweighted counts, the result is integer. For weighted counts, the result is float. If cumulative is False, result[i] contains the counts with (-inf if i == 0 else r[i-1]) < R <= r[i]

Notes

Pair-counting is the basic operation used to calculate the two point correlation functions from a data set composed of position of objects.

Two point correlation function measures the clustering of objects and is widely used in cosmology to quantify the large scale structure in our Universe, but it may be useful for data analysis in other fields where self-similar assembly of objects also occur.

The Landy-Szalay estimator for the two point correlation function of D measures the clustering signal in D. [2]

For example, given the position of two sets of objects,

• objects D (data) contains the clustering signal, and

• objects R (random) that contains no signal,

\[\xi(r) = \frac{<D, D> - 2 f <D, R> + f^2<R, R>}{f^2<R, R>},\]

where the brackets represents counting pairs between two data sets in a finite bin around r (distance), corresponding to setting cumulative=False, and f = float(len(D)) / float(len(R)) is the ratio between number of objects from data and random.

The algorithm implemented here is loosely based on the dual-tree algorithm described in [1]. We switch between two different pair-cumulation scheme depending on the setting of cumulative. The computing time of the method we use when for cumulative == False does not scale with the total number of bins. The algorithm for cumulative == True scales linearly with the number of bins, though it is slightly faster when only 1 or 2 bins are used. [5].

As an extension to the naive pair-counting, weighted pair-counting counts the product of weights instead of number of pairs. Weighted pair-counting is used to estimate marked correlation functions ([3], section 2.2), or to properly calculate the average of data per distance bin (e.g. [4], section 2.1 on redshift).

1(1,2)

Gray and Moore, “N-body problems in statistical learning”, Mining the sky, 2000, https://arxiv.org/abs/astro-ph/0012333

2

Landy and Szalay, “Bias and variance of angular correlation functions”, The Astrophysical Journal, 1993, http://adsabs.harvard.edu/abs/1993ApJ…412…64L

3

Sheth, Connolly and Skibba, “Marked correlations in galaxy formation models”, Arxiv e-print, 2005, https://arxiv.org/abs/astro-ph/0511773

4

Hawkins, et al., “The 2dF Galaxy Redshift Survey: correlation functions, peculiar velocities and the matter density of the Universe”, Monthly Notices of the Royal Astronomical Society, 2002, http://adsabs.harvard.edu/abs/2003MNRAS.346…78H

5

https://github.com/scipy/scipy/pull/5647#issuecomment-168474926

data

indices

leafsize

m

maxes

mins

n

query(self, x, k=1, eps=0, p=2, distance_upper_bound=np.inf, n_jobs=1)

Query the kd-tree for nearest neighbors

Parameters

xarray_like, last dimension self.m

An array of points to query.

klist of integer or integer

The list of k-th nearest neighbors to return. If k is an integer it is treated as a list of [1, … k] (range(1, k+1)). Note that the counting starts from 1.

epsnon-negative float

Return approximate nearest neighbors; the k-th returned value is guaranteed to be no further than (1+eps) times the distance to the real k-th nearest neighbor.

pfloat, 1<=p<=infinity

Which Minkowski p-norm to use. 1 is the sum-of-absolute-values “Manhattan” distance 2 is the usual Euclidean distance infinity is the maximum-coordinate-difference distance A finite large p may cause a ValueError if overflow can occur.

distance_upper_boundnonnegative float

Return only neighbors within this distance. This is used to prune tree searches, so if you are doing a series of nearest-neighbor queries, it may help to supply the distance to the nearest neighbor of the most recent point.

n_jobsint, optional

Number of jobs to schedule for parallel processing. If -1 is given all processors are used. Default: 1.

Returns

darray of floats

The distances to the nearest neighbors. If x has shape tuple+(self.m,), then d has shape tuple+(k,). When k == 1, the last dimension of the output is squeezed. Missing neighbors are indicated with infinite distances.

indarray of ints

The locations of the neighbors in self.data. If x has shape tuple+(self.m,), then i has shape tuple+(k,). When k == 1, the last dimension of the output is squeezed. Missing neighbors are indicated with self.n.

Notes

If the KD-Tree is periodic, the position x is wrapped into the box.

When the input k is a list, a query for arange(max(k)) is performed, but only columns that store the requested values of k are preserved. This is implemented in a manner that reduces memory usage.

Examples

>>> import numpy as np
>>> from scipy.spatial import cKDTree
>>> x, y = np.mgrid[0:5, 2:8]
>>> tree = cKDTree(np.c_[x.ravel(), y.ravel()])

To query the nearest neighbours and return squeezed result, use

>>> dd, ii = tree.query([[0, 0], [2.1, 2.9]], k=1)
>>> print(dd, ii)
[2.         0.14142136] [ 0 13]

To query the nearest neighbours and return unsqueezed result, use

>>> dd, ii = tree.query([[0, 0], [2.1, 2.9]], k=[1])
>>> print(dd, ii)
[[2.        ]
 [0.14142136]] [[ 0]
 [13]]

To query the second nearest neighbours and return unsqueezed result, use

>>> dd, ii = tree.query([[0, 0], [2.1, 2.9]], k=[2])
>>> print(dd, ii)
[[2.23606798]
 [0.90553851]] [[ 6]
 [12]]

To query the first and second nearest neighbours, use

>>> dd, ii = tree.query([[0, 0], [2.1, 2.9]], k=2)
>>> print(dd, ii)
[[2.         2.23606798]
 [0.14142136 0.90553851]] [[ 0  6]
 [13 12]]

or, be more specific

>>> dd, ii = tree.query([[0, 0], [2.1, 2.9]], k=[1, 2])
>>> print(dd, ii)
[[2.         2.23606798]
 [0.14142136 0.90553851]] [[ 0  6]
 [13 12]]

query_ball_point(self, x, r, p=2., eps=0)

Find all points within distance r of point(s) x.

Parameters

xarray_like, shape tuple + (self.m,)

The point or points to search for neighbors of.

rarray_like, float

The radius of points to return, shall broadcast to the length of x.

pfloat, optional

Which Minkowski p-norm to use. Should be in the range [1, inf]. A finite large p may cause a ValueError if overflow can occur.

epsnonnegative float, optional

Approximate search. Branches of the tree are not explored if their nearest points are further than r / (1 + eps), and branches are added in bulk if their furthest points are nearer than r * (1 + eps).

n_jobsint, optional

Number of jobs to schedule for parallel processing. If -1 is given all processors are used. Default: 1.

return_sortedbool, optional

Sorts returned indicies if True and does not sort them if False. If None, does not sort single point queries, but does sort multi-point queries which was the behavior before this option was added.

New in version 1.2.0.

return_length: bool, optional

Return the number of points inside the radius instead of a list of the indices. .. versionadded:: 1.3.0

Returns

resultslist or array of lists

If x is a single point, returns a list of the indices of the neighbors of x. If x is an array of points, returns an object array of shape tuple containing lists of neighbors.

Notes

If you have many points whose neighbors you want to find, you may save substantial amounts of time by putting them in a cKDTree and using query_ball_tree.

Examples

>>> from scipy import spatial
>>> x, y = np.mgrid[0:4, 0:4]
>>> points = np.c_[x.ravel(), y.ravel()]
>>> tree = spatial.cKDTree(points)
>>> tree.query_ball_point([2, 0], 1)
[4, 8, 9, 12]

query_ball_tree(self, other, r, p=2., eps=0)

Find all pairs of points whose distance is at most r

Parameters

othercKDTree instance

The tree containing points to search against.

rfloat

The maximum distance, has to be positive.

pfloat, optional

Which Minkowski norm to use. p has to meet the condition 1 <= p <= infinity. A finite large p may cause a ValueError if overflow can occur.

epsfloat, optional

Approximate search. Branches of the tree are not explored if their nearest points are further than r/(1+eps), and branches are added in bulk if their furthest points are nearer than r * (1+eps). eps has to be non-negative.

Returns

resultslist of lists

For each element self.data[i] of this tree, results[i] is a list of the indices of its neighbors in other.data.

query_pairs(self, r, p=2., eps=0)

Find all pairs of points whose distance is at most r.

Parameters

rpositive float

The maximum distance.

pfloat, optional

Which Minkowski norm to use. p has to meet the condition 1 <= p <= infinity. A finite large p may cause a ValueError if overflow can occur.

epsfloat, optional

Approximate search. Branches of the tree are not explored if their nearest points are further than r/(1+eps), and branches are added in bulk if their furthest points are nearer than r * (1+eps). eps has to be non-negative.

output_typestring, optional

Choose the output container, ‘set’ or ‘ndarray’. Default: ‘set’

Returns

resultsset or ndarray

Set of pairs (i,j), with i < j, for which the corresponding positions are close. If output_type is ‘ndarray’, an ndarry is returned instead of a set.

size

sparse_distance_matrix(self, other, max_distance, p=2.)

Compute a sparse distance matrix

Computes a distance matrix between two cKDTrees, leaving as zero any distance greater than max_distance.

Parameters

othercKDTree

max_distancepositive float

pfloat, 1<=p<=infinity

Which Minkowski p-norm to use. A finite large p may cause a ValueError if overflow can occur.

output_typestring, optional

Which container to use for output data. Options: ‘dok_matrix’, ‘coo_matrix’, ‘dict’, or ‘ndarray’. Default: ‘dok_matrix’.

Returns

resultdok_matrix, coo_matrix, dict or ndarray

Sparse matrix representing the results in “dictionary of keys” format. If a dict is returned the keys are (i,j) tuples of indices. If output_type is ‘ndarray’ a record array with fields ‘i’, ‘j’, and ‘v’ is returned,

tree