def radius_outlier_filter(input_cloud, r=1, search_eps=0, p=2, sd_cutoff=1):
output_cloud = []
print("Constructing kdtree")
tree = kdtree.KDTree(input_cloud)
other_tree = kdtree.KDTree(input_cloud)
print("kdtree constructed")
print("finding neighbors")
# neighbor_list = []
# for point in tqdm(input_cloud, total=len(input_cloud), desc="Finding Neighbors"):
# neighbor_list.append(tree.query_ball_point(point, r=r, eps=search_eps))
start = time.time()
neighbor_list = tree.query_ball_tree(other_tree, r=r, p=p, eps=search_eps)
end = time.time() - start
print("Finding neighbors took %s seconds" % end)
lengths = [len(x) for x in neighbor_list]
mean = np.mean(lengths)
std = np.std(lengths)
cutoff = mean - sd_cutoff * std
for i in trange(len(neighbor_list), desc="Removing outliers"):
if lengths[i] >= cutoff:
output_cloud.append(input_cloud[i])
return output_cloud
def __Debounce(self, xs, ys, radius=4):
if len(xs) <= 1:
return xs, ys
kdt = kdtree.KDTree(scipy.array([xs, ys]).T)
xsd = []
ysd = []
for xi, yi in zip(xs, ys):
neigh = kdt.query_ball_point([xi, yi], radius)
if len(neigh) > 1:
Ii = self.filteredData[xi, yi]
In = self.filteredData[xs[neigh].astype('i'),
ys[neigh].astype('i')].max()
if not Ii < In:
xsd.append(xi)
ysd.append(yi)
else:
xsd.append(xi)
ysd.append(yi)
return xsd, ysd
def smooth(self, x):
"""
Apply K Nearest Neighbors density estimator over a grid.
Arguments:
1. x: 2D dataset to be smoothed. It is assumed that the rows of
the data matrix are the sample points.
Returns:
1. Smoothed function over the specified domain.
Example:
TODO: Write sample code...
"""
domx, domy = self._construct_domain()
dom = np.vstack((domx.ravel(), domy.ravel())).T
# construct KD tree on data
tree = kd.KDTree(x)
# get k^{th} nearest neighbors to each point in the domain
dist = tree.query(dom, k=self.k, p=2)[0]
dist_knn = dist[:, self.k - 1].reshape(self.domy_params[2],
self.domx_params[2])
dist_knn = np.divide(self.k / (x.shape[0] * np.pi), dist_knn**2)
# KNN density estimator
return (dist_knn)
def iteration(self):
log.info('clusters initialised ...')
# initialise clusters
self.initialise_clusters()
log.info('clusters moved briefly ...')
# move clusters based on gradiant magnitude value
self.initial_move_clusters()
log.info('main process started ...')
for i in trange(self.max_iteration):
# assign pixels to closest cluster based on total distance value
self.assign_pixels_to_clusters()
# update cluster centers based on new pixel to cluster array
self.update_cluster_centers()
log.info('main process done ...')
if self.post_process:
log.info('connected components started ...')
# calculate kd_tree data
centers = [[cluster.w, cluster.h] for cluster in self.clusters]
self.kd_tree = kd.KDTree(centers)
for i in trange(1):
self.connected_component_post_process()
log.warning('connected components done ...')
# compute segmented image
self.compute_segmented_image()
def fitPts(pts, NFits = 0):
xm, ym, zm = pts.min(0)
xmx, ymx, zmx = pts.max(0)
X, Y, Z = np.mgrid[xm:xmx:5, ym:ymx:5, zm:zmx:5]
im = np.zeros(X.shape)
kdt = kdtree.KDTree(pts)
if NFits == 0:
NFits = pts.shape[0]
for i in np.arange(NFits):
ptr = pts[kdt.query_ball_point(pts[i, :], 150),:]
#po = mean(ptr, 0)
po = pts[i, :]
if ptr.shape[0] > 10:
ptr = ptr - po[None, :]
ps = []
mfs = []
for i in range(5):
op = leastsq(mf, list(1 + .1*np.random.normal(size=9)) + [-1], args=(ptr[:,0],ptr[:,1],ptr[:,2]), full_output=1, epsfcn=1)
ps.append(op[0])
mfs.append((op[2]['fvec']**2).sum())
#print mfs
p3 = ps[np.argmin(mfs)]
fv = rend(p3, po, 100, X, Y, Z, im, xm, ym, zm, 5,5,5, 1.0/ptr.shape[0])
return im
def learn(self,learndataset):
"""learn the KNN structure required to evaluate new instances
:param learndataset: learning instances
:type learndataset: :class:`~classifip.dataset.arff.ArffFile`
"""
self.__init__()
self.classes=learndataset.attribute_data['class'][:]
#Initialize average distance for every possible class
for i in learndataset.attribute_data['class']:
class_set=learndataset.select_class([i])
values=[row[0:len(row)-1] for row in class_set.data]
if len(values) > 1000:
valred=np.random.permutation(values)[0:1000]
class_distances=distance.cdist(valred,valred)
else:
class_distances=distance.cdist(values,values)
average=class_distances.sum()/(len(class_distances)**2
-len(class_distances))
self.av_dist.append(average)
# training the whole thing
learndata=[row[0:len(row)-1] for row in learndataset.data]
self.truelabels=[row[-1] for row in learndataset.data]
self.tree=kdtree.KDTree(learndata)
def nearest_boundaries_location(self, x, y, xi, yi):
"""
This method find the nearest points between the mother grid and the boundaries of the nested grid.
"""
mdgrd = np.array(zip(x.ravel(), y.ravel()))
kdt = kd.KDTree(mdgrd)
self.mdi_dist, self.mdi = kdt.query(np.array(zip(xi, yi)))
def __init__(self, vert):
self.num_vert = vert
self.bowlr = 2.1 * pow(
log(self.num_vert) / float(self.num_vert), 1 / float(NUM_DIM))
self.delta = 0.001
self.nodes = []
self.point = []
self.tree = []
self.mydict = []
self.truth = []
self.observations = []
self.estimates = []
for i in range(self.num_vert):
self.nodes.append(Node(False))
# initial variance
for i in range(self.num_vert):
n1 = self.nodes[i]
n1.density = normal_val(n1.x, array(init_state), init_var)
self.normalize_density()
self.points = [self.key(mynode.x) for mynode in self.nodes]
self.tree = kdtree.KDTree(self.points)
def learn(self,learndataset,likovo_normalise=True):
"""learn the tree structure required to perform evaluation
:param learndataset: learning instances
:type learndataset: :class:`~classifip.dataset.arff.ArffFile`
:param likovo_normalise: normalise the input features or not
:type likovo_normalise: boolean
"""
self.classes=learndataset.attribute_data['class'][:]
learndata=[row[0:len(row)-1] for row in learndataset.data]
data_array=np.array(learndata).astype(float)
if likovo_normalise == True:
self.normal.append(True)
span=data_array.max(axis=0)-data_array.min(axis=0)
self.normal.append(span)
self.normal.append(data_array.min(axis=0))
data_array=(data_array-data_array.min(axis=0))/span
else:
self.normal.append(False)
#Initalise radius as average distance between all learning instances
if len(data_array) > 1000:
valred=np.random.permutation(data_array)[0:1000]
distances=distance.cdist(valred,valred)
else:
distances=distance.cdist(data_array,data_array)
self.radius=distances.sum()/(2*(len(distances)**2-len(distances)))
self.tree=kdtree.KDTree(data_array)
self.trueclasses=[row[-1] for row in learndataset.data]
def lines_distance(lines, r=2.):
"""
2 x 3
:param lines:
:return:
"""
tree = kdtree.KDTree(np.concatenate(lines))
tree.query_pairs(r)
def kdtree(self):
if not hasattr(self, '_kdtree'):
self._kdtree = None
if self._kdtree is None:
mat = self.__create_numpy_matrix(self.graph)
self._kdtree = kdtree.KDTree(mat[:, iLoc.X:iLoc.Z + 1])
return self._kdtree
def hybrid_astar_planning(sx, sy, syaw, gx, gy, gyaw, ox, oy, xyreso, yawreso):
sxr, syr = round(sx / xyreso), round(sy / xyreso)
gxr, gyr = round(gx / xyreso), round(gy / xyreso)
syawr = round(rs.pi_2_pi(syaw) / yawreso)
gyawr = round(rs.pi_2_pi(gyaw) / yawreso)
nstart = Node(sxr, syr, syawr, 1, [sx], [sy], [syaw], [1], 0.0, 0.0, -1)
ngoal = Node(gxr, gyr, gyawr, 1, [gx], [gy], [gyaw], [1], 0.0, 0.0, -1)
kdtree = kd.KDTree([[x, y] for x, y in zip(ox, oy)])
P = calc_parameters(ox, oy, xyreso, yawreso, kdtree)
hmap = astar.calc_holonomic_heuristic_with_obstacle(
ngoal, P.ox, P.oy, P.xyreso, 1.0)
steer_set, direc_set = calc_motion_set()
open_set, closed_set = {calc_index(nstart, P): nstart}, {}
qp = QueuePrior()
qp.put(calc_index(nstart, P), calc_hybrid_cost(nstart, hmap, P))
while True:
if not open_set:
return None
ind = qp.get()
n_curr = open_set[ind]
closed_set[ind] = n_curr
open_set.pop(ind)
update, fpath = update_node_with_analystic_expantion(n_curr, ngoal, P)
if update:
fnode = fpath
break
for i in range(len(steer_set)):
node = calc_next_node(n_curr, ind, steer_set[i], direc_set[i], P)
if not node:
continue
node_ind = calc_index(node, P)
if node_ind in closed_set:
continue
if node_ind not in open_set:
open_set[node_ind] = node
qp.put(node_ind, calc_hybrid_cost(node, hmap, P))
else:
if open_set[node_ind].cost > node.cost:
open_set[node_ind] = node
qp.put(node_ind, calc_hybrid_cost(node, hmap, P))
return extract_path(closed_set, fnode, nstart)
def guided_filter_kNN(input_cloud, k=50, filter_eps=.05):
output_cloud = []
# put all points in kdtree
print("Constructing kdtree")
tree = kdtree.KDTree(input_cloud)
# other_tree = kdtree.KDTree(input_cloud)
print("kdtree constructed")
print("finding neighbors")
neighbor_list = []
for q in tqdm(input_cloud,
total=len(input_cloud),
desc="Querying neighbors"):
_, neighbors = tree.query(
x=q, k=k, eps=0.5
) # this eps allows for a little inaccuracy for speed tradeoff
neighbor_list.append(neighbors)
# neighbor_list = tree.query(x=input_cloud, k=k, distance_upper_bound=5)
# print(neighbor_list)
print("neighbors found")
# for point in tqdm(input_cloud, total=len(input_cloud)):
# neighbors = tree.query_ball_point(point, radius)
for i in trange(len(input_cloud), desc="Filtering"):
# remember that neighbor list gives a list of indices that need to be pulled from input_list
# step 1, as referred to in the paper
neighbors = neighbor_list[i]
# step 2
k = float(len(neighbors))
# step 3
p_bar = np.asarray([0., 0., 0.])
for n in neighbors:
# print(input_cloud[n])
try:
p_bar += input_cloud[n]
except IndexError:
pdb.set_trace()
p_bar /= k
# print(p_bar)
# step 4
temp = 0
for n in neighbors:
temp += np.dot(input_cloud[n], input_cloud[n])
temp /= k
centroid_dot = np.dot(p_bar, p_bar)
a = (temp - centroid_dot) / ((temp - centroid_dot) + filter_eps)
# step 5
b = p_bar - a * p_bar
# step 6
output_cloud.append(a * input_cloud[i] + b)
return output_cloud
def __init__(self, points, step_size=30, tolerance=50):
self.step_size = step_size
np.random.seed(0)
self.unsorted = np.array(points)
self.n = len(points)
self.sortedxs = []
self.sortedys = []
self.kdtree = kdtree.KDTree(points)
self.tolerance = tolerance
initpoint = self.unsorted[self.n / 2]
i = self.step_size
while True:
i = i * 2
neighbors = self.kdtree.query_ball_point(initpoint, i)
if len(neighbors) >= self.tolerance or i > self.step_size * 3:
break
pca = PCA(n_components=2)
pca.fit(self.unsorted[neighbors])
initdirection1 = pca.components_[0]
initdirection2 = -initdirection1
p = initpoint
d = initdirection1
oldp = p
while p is not None:
p, d = self.searchnext(p, d)
if p is None or (oldp[0] == p[0] and oldp[1] == p[1]):
break
oldp = p
if p is not None:
arr1 = np.zeros(1)
arr2 = np.zeros(1)
arr1[0] = p[0]
arr2[0] = p[1]
if len(self.sortedxs) == 0:
self.sortedxs = arr1
self.sortedys = arr2
else:
self.sortedxs = np.concatenate((arr1, self.sortedxs))
self.sortedys = np.concatenate((arr2, self.sortedys))
self.sortedxs = np.append(self.sortedxs, initpoint[0])
self.sortedys = np.append(self.sortedys, initpoint[1])
p = initpoint
d = initdirection2
oldp = p
while p is not None:
p, d = self.searchnext(p, d)
if p is None or (oldp[0] == p[0] and oldp[1] == p[1]):
break
oldp = p
if p is not None:
self.sortedxs = np.append(self.sortedxs, p[0])
self.sortedys = np.append(self.sortedys, p[1])
self.sorted = np.transpose(np.array([self.sortedxs, self.sortedys]))
def __init__(self, data, search_type='brute-force'):
self.__data = data
self.__nr, self.__nc = data.shape
if search_type == 'brute-force':
self.__search_type = search_type
elif search_type == 'kdtree':
self.__search_type = search_type
self.__kdtree = kdtree.KDTree(data)
elif search_type == 'mykdtree':
self.__search_type = search_type
self.__kdtree = MyKDTree(data)
else:
raise Exception('unknown search type: {}'.format(search_type))
def learn(self, learn_data_set, nb_labels, learn_disc_set=None):
"""
Warning: it should be normalized since it do an Euclidean distance
"""
self.__init__()
self.nb_labels = nb_labels
self.nda_models = dict()
self.nb_feature = len(learn_data_set.attributes[:-self.nb_labels])
_np_data = np.array(learn_data_set.data)
_index_features = np.array(range(self.nb_feature))
_index_labels = np.array(
np.arange(self.nb_feature, self.nb_feature + self.nb_labels))
self.x_learning = np.array(_np_data[:, _index_features],
dtype=np.float)
self.y_learning = _np_data[:, _index_labels].copy()
# procedure create kd_tree by classifier with missing instances
self.kd_tree = dict()
self.radius = np.ones(nb_labels)
for label_index in range(nb_labels):
missing_index = []
# assuming the index learn_disc_set and learn_data_set are same (salmuz)
for row_index, row_instance in enumerate(learn_disc_set.data):
if row_instance[self.nb_feature + label_index] == '-1':
missing_index.append(row_index)
x_marginal = np.delete(self.x_learning, missing_index, axis=0)
_distances = distance.cdist(x_marginal, self.x_learning)
self.radius[label_index] = np.mean(_distances[np.tril_indices(
len(x_marginal), k=-1)])
self.kd_tree[label_index] = kdtree.KDTree(x_marginal)
self.learn_disc_set = learn_disc_set.make_clone()
# vacuous skeleton arff file
self.skeleton_learn_knn = ArffFile()
self.skeleton_learn_knn.attribute_data = self.learn_disc_set.attribute_data.copy(
)
self.skeleton_learn_knn.attribute_types = self.learn_disc_set.attribute_types.copy(
)
self.skeleton_learn_knn.attributes = self.learn_disc_set.attributes.copy(
)
for label_name in learn_data_set.attributes[-self.nb_labels:]:
del self.skeleton_learn_knn.attribute_data[label_name]
del self.skeleton_learn_knn.attribute_types[label_name]
self.skeleton_learn_knn.attributes.pop()
self.skeleton_learn_knn.data = list()
self.skeleton_learn_knn.define_attribute(name="class",
atype="nominal",
data=['0', '1'])
def generate(self, n, bounds=DobotModel.limits):
"""
Generates (or regenerates) the PRM given a target number of samples n
"""
self.G = nx.Graph()
# Sample environment
ps = self._sample_cs(n, bounds)
# ps = self._sample_ws(n,np.array([[0,300],[-200,200],[0,200]]))
self.tree = kdt.KDTree(ps)
# Connect samples
for k in xrange(self.tree.n):
self._connect(k, self.tree.data[k])
def cleanup(self):
"""Call after accretion; merges failed bins into neighbours"""
self._valid_bins = np.array(self._valid_bins, dtype=np.bool)
self._current_bin_centroids = np.array(self._current_bin_centroids)
good_bins = np.where(self._valid_bins == True)[0] # NOQA
failed_bins = np.where(self._valid_bins == False)[0] # NOQA
# build a kdtree of good bins
tree = kdtree.KDTree(self._current_bin_centroids[good_bins, :])
for i, failed_idx in enumerate(failed_bins):
# update the segmentation image
pix_idx = np.where(self._seg_image == failed_idx)
# self._seg_image[pix_idx] = -1
coords = np.vstack(pix_idx).T
dists, reassignment_indices = tree.query(coords)
self._seg_image[pix_idx] = good_bins[reassignment_indices]
def compute_normals_for_all_points(points, n_size=36):
"""
:param n_size:
:param points:
:return:
"""
tree = kdtree.KDTree(points)
normals = []
for point in points:
d, i = tree.query(point, k=n_size)
current_points = points[i] - point
normals.append(compute_normal_from_points(current_points))
return np.array(normals)
def smooth(self, x):
k = np.ceil(self.tau * x.shape[0]) if self.tau is not None else self.k
k = int(k)
domx, domy = self._construct_domain()
dom = np.vstack((domx.ravel(), domy.ravel())).T
tree = kd.KDTree(x)
knn = tree.query(dom, k=k, p=2)[1]
#
Fxy = np.subtract(dom[:, np.newaxis, :], x[knn, :]).reshape(-1, 2)
Fxy = np.linalg.norm(Fxy, axis=1).reshape(-1, k)
Fxy = np.sqrt(Fxy.mean(axis=1))
return (Fxy.reshape(self.domy_params[2], self.domx_params[2]))
def BridgeSubgraphs(self, graph, ANodes, BNodes):
'''Create edges between the closest locations of isolated subgraphs'''
matA = self.__create_numpy_matrix(ANodes)
matB = self.__create_numpy_matrix(BNodes)
AKDTree = kdtree.KDTree(matA[:, iLoc.X:iLoc.Z + 1])
distances = AKDTree.query(matB[:, iLoc.X:iLoc.Z + 1], k=1)
index_of_minB = distances[0].argmin()
index_of_minA = distances[1][index_of_minB]
NodeA = ANodes[index_of_minA]
NodeB = BNodes[index_of_minB]
graph.add_edge(NodeA, NodeB)
def __init__(self, vert):
self.num_vert = vert
self.bowlr = 2.1 * pow(
log(self.num_vert) / float(self.num_vert), 1 / float(NUM_DIM))
self.delta = 0.1
self.nodes = []
self.point = []
self.tree = []
self.mydict = []
self.nodes.append(Node(True))
for i in range(self.num_vert - 1):
self.nodes.append(Node())
self.points = [self.key(mynode.x) for mynode in self.nodes]
self.tree = kdtree.KDTree(self.points)
def get_edges_in_sphere(G,
centerSphere,
radiusSphere,
nkinds,
radiusSphereMin=0):
""" Returns a list of edges which lie in the specified sphere. The function
uses the tortuous vessel properties (if one of the points of the vessel is
located inside the sphere the vessel is considered to be in the sphere)
INPUT: G: main Graph
centerSphere: the coordinates of the center of the sphere
radiusSphere: the radius of the center of the sphere
nkinds: list of vessel kinds which should be considered
radiusSphereMin: default = 0, a value between [0,radiusSphere] can be
given. vessels inside the sphereMin are not considered
OUTPUT: edges: list of edges where at least one point along the edge is
located inside the sphere
"""
#Get tortuous values of all edges of interest
rAll = []
edgeAll = []
for nkind in nkinds:
for e in G.es(nkind_eq=nkind):
for i in e['points']:
rAll.append(i)
edgeAll.append(e.index)
Kdt = kdtree.KDTree(rAll, leafsize=10)
kAll = 100
nearestAll = Kdt.query(centerSphere, k=kAll)
nearestDist = nearestAll[0]
nearestIndex = nearestAll[1]
largestDist = nearestDist[-1]
while largestDist < radiusSphere:
kAll += 100
nearestAll = Kdt.query(centerSphere, k=kAll)
nearestDist = nearestAll[0]
nearestIndex = nearestAll[1]
largestDist = nearestDist[-1]
edges = []
for i, j in zip(nearestIndex, nearestDist):
if edgeAll[
i] not in edges and j < radiusSphere and j >= radiusSphereMin:
edges.append(edgeAll[i])
return edges
def connect_vasculature_with_tissue(self, **kwargs):
"""Connects VascularGraphs G1 (vasculature) and G2 (tissue) with
traited 'soft links'. Each vascualar vertex connects to exactly one
tissue vertex (its nearest neighbor). A tissue vertex, however, may
connect to multiple vascular vertices. Each connection also stores
the sum of the products of surface area and exchange coefficient. This
vertex property is required for the exchange of a given substance
between vasculature and tissue (see the Exchange class).
INPUT: **kwargs
substance: The name of the substance that is to be exchanged.
OUTPUT: None
"""
self.vertexG1ToVerticesG2 = {}
self.vertexG2ToVerticesG1 = {}
self.connections = []
G1 = self._G1
G2 = self._G2
substance = kwargs['substance']
Kdt = kdtree.KDTree(G2.vs['r'], leafsize=10)
for vIndex in xrange(G1.vcount()):
edgeIndices = G1.adjacent(vIndex, 'all')
exchangeFactor = 0.0
for edge in G1.es(edgeIndices):
exchangeFactor = exchangeFactor + edge['length'] * np.pi * \
edge['diameter'] / 2.0 * \
edge['exchangeCoefficient'][substance]
if 'kind' in G1.vs.attribute_names():
if G1.vs[vIndex]['kind'] == 'u':
exchangeFactor = exchangeFactor + \
G1.vs[vIndex]['uSurfaceArea'] * \
G1.vs[vIndex]['uExchangeCoefficient'][substance]
tissueNeighbor = int(Kdt.query(G1.vs[vIndex]['r'])[1])
self.vertexG1ToVerticesG2[vIndex] = tissueNeighbor
if self.vertexG2ToVerticesG1.has_key(tissueNeighbor):
self.vertexG2ToVerticesG1[tissueNeighbor].append(vIndex)
else:
self.vertexG2ToVerticesG1[tissueNeighbor] = [vIndex]
self.connections.append(self.Connection(vIndex, tissueNeighbor))
self.connections[-1].exchangeFactor = exchangeFactor
def classify_close_by(points,
labels,
from_label,
to_label,
close_to_label,
radius,
kdTree=None):
"""
:param radius:
:param close_to_label:
:param to_label:
:param from_label:
:param labels:
:param points:
:param kdTree:
:return:
"""
if kdTree is None:
kdTree = kdtree.KDTree(points)
for i, point_label in enumerate(zip(points, labels)):
point = point_label[0]
label = point_label[1]
if label != from_label:
continue
i = kdTree.query_ball_point(point, r=radius)
qpoints = points[i]
qlabels = labels[i]
npoints = qpoints[qlabels == close_to_label]
if len(npoints) == 0:
continue
labels[i] = to_label
return labels
def smooth(self, x):
"""
Apply K Nearest Neighbors adaptive Gaussian smoother to a provided 2D dataset.
Arguments:
1. x: 2D dataset to be smoothed. It is assumed that the rows of
the data matrix are the sample points.
Returns:
1. Smoothed function over the specified domain.
Example:
TODO: Write sample code...
"""
domx, domy = self._construct_domain()
dom = np.vstack((domx.ravel(), domy.ravel())).T
# construct KD tree on data
tree = kd.KDTree(x)
# get k nearest neighbors
dist = tree.query(dom, k=self.k)[0]
tp_knn = dist[:, self.k - 1].reshape(-1, 1)
### ADAPTIVE KERNEL SMOOTHING
# pairwise subtraction between grid points and each data point
# reshape from tensor to matrix (K x 2)
Fxy = np.subtract(dom[:, np.newaxis, :],
x[np.newaxis, :, :]).reshape(-1, 2)
Fxy = np.square(np.linalg.norm(Fxy, axis=1)).reshape(dom.shape[0], -1)
Fxy = np.divide(Fxy, -2 * tp_knn**2)
Fxy = np.divide(np.exp(Fxy), 2 * np.pi * tp_knn**2)
Fxy = Fxy.mean(axis=1)
return (Fxy.reshape(self.domy_params[2], self.domx_params[2]))
def learn(self, learndataset, pipp_normalise=True):
"""learn the tree structure required to perform evaluation
:param learndataset: learning instances
:type learndataset: :class:`~classifip.dataset.arff.ArffFile`
:param pipp_normalise: normalise the input features or not
:type pipp_normalise: boolean
.. note::
learndataset should come from a xarff file tailored for lable ranking
"""
self.labels = learndataset.attribute_data['L'][:]
learndata = [row[0:len(row) - 1] for row in learndataset.data]
data_array = np.array(learndata).astype(float)
if pipp_normalise == True:
span = data_array.max(axis=0) - data_array.min(axis=0)
self.normal.append(True)
self.normal.append(span)
self.normal.append(data_array.min(axis=0))
data_array = (data_array - data_array.min(axis=0)) / span
else:
self.normal.append(False)
#Initalise radius as average distance between all learning instances
if len(data_array) > 1000:
data_red = np.random.permutation(data_array)[0:1000]
distances = distance.cdist(data_red, data_red)
else:
distances = distance.cdist(data_array, data_array)
self.radius = distances.sum() / (2 *
(len(distances)**2 - len(distances)))
self.tree = kdtree.KDTree(data_array)
self.truerankings = [
ranking_matrices(row[-1], self.labels) for row in learndataset.data
]
def __init__(self, vert):
self.num_vert = vert
self.bowlr = 2.1*pow(log(self.num_vert)/float(self.num_vert),1/float(NUM_DIM))
self.delta = 0.01
self.nodes = []
self.edges = []
self.tree = []
#mydict = (key = node, data = [ [in_edge_1, ... ], [ out_edge_1, ...]])
self.mydict = dict()
for i in range(self.num_vert):
n1 = Node()
self.nodes.append(n1)
# initial variance
for i in range(self.num_vert):
n1 = self.nodes[i]
n1.density = normal_val(n1.x, array(init_state), init_var)
self.normalize_density()
self.points = [self.key(mynode.x) for mynode in self.nodes]
self.tree = kdtree.KDTree(self.points)
def vertices_from_coordinates(G,
coordinates,
diameter_ll=0.0,
isEndpoint=False):
"""Given a list of x,y,z coordinates, locate the most closely matching set
of verties in a vascular graph of iGraph format.
INPUT: G: Vascular graph in iGraph format.
coordinates: List of lists specifying the coordinates (i.e.:
[[x1,y1,z1],[x2,y2,z2],...]).
diameter_ll: (Optional) lower limit of edge diameter, to select only
those vertices bordering a sufficiently large diameter
edge. Default is 0.0, i.e. all vertices are considered.
isEndpoint: Boolean whether or not the vertex searched for is
required to be an endpoint. Default is 'False'.
OUTPUT: vertex_indices: Array of vertex indices that represent the best
matches.
distances: Array of distances that the best matching vertices are
separated from the supplied coordinates. Units match
those of the graph vertex coordinates.
"""
# Select vertex indices based on diameter of adjacent edges:
si = unique(
flatten([G.es[x].tuple
for x in G.es(diameter_ge=diameter_ll).indices])).tolist()
# Optionally filter for end-points:
if isEndpoint:
si = [i for i in si if G.degree(i) == 1]
# Construct k-dimensional seach-tree:
kdt = kdtree.KDTree(G.vs[si]['r'], leafsize=10)
search_result = kdt.query(coordinates)
sr_v = np.ravel([search_result[1]]).tolist()
vertex_indices = [si[x] for x in sr_v]
distances = np.ravel([search_result[0]]).tolist()
return vertex_indices, distances
def get_kdtree(self):
"""Build a mocked KD-Tree."""
all_foodtrucks, points = [], []
cur_dir = os.path.dirname(os.path.realpath(__file__))
dir = cur_dir + '/fixtures/foodtrucks.json'
with open(dir, 'r') as f:
all_foodtrucks = json.load(f)
for foodtruck in all_foodtrucks:
if 'x' in foodtruck and 'y' in foodtruck:
try:
x = float(foodtruck['x'])
except ValueError:
raise InvalidValueError('x', foodtruck['x'])
try:
y = float(foodtruck['y'])
except ValueError:
raise InvalidValueError('y', foodtruck['y'])
point = (x, y)
points.append(point)
kd_tree = kdtree.KDTree(points)
return kd_tree