def _masked_array_type_from_col(col):
"""
Return a type representing a tuple of arrays,
the first element an array of the numba type
corresponding to `dtype`, and the second an
array of bools representing a mask.
"""
nb_scalar_ty = numpy_support.from_dtype(col.dtype)
if col.mask is None:
return nb_scalar_ty[::1]
else:
return Tuple((nb_scalar_ty[::1], libcudf_bitmask_type[::1]))
def _construct_signature(frame, return_type, args):
"""
Build the signature of numba types that will be used to
actually JIT the kernel itself later, accounting for types
and offsets. Skips columns with unsupported dtypes.
"""
# Tuple of arrays, first the output data array, then the mask
return_type = Tuple((return_type[::1], boolean[::1]))
offsets = []
sig = [return_type, int64]
for col in _supported_cols_from_frame(frame).values():
sig.append(_masked_array_type_from_col(col))
offsets.append(int64)
# return_type, size, data, masks, offsets, extra args
sig = void(*(sig + offsets + [typeof(arg) for arg in args]))
return sig
def create_signature(retty, args):
"""
Given the return type and arguments for a libdevice function, return the
signature of the stub function used to call it from CUDA Python.
"""
# Any pointer arguments should be part of the return type.
return_types = [arg.ty for arg in args if arg.is_ptr]
# If the return type is void, there is no point adding it to the list of
# return types.
if retty != void:
return_types.insert(0, retty)
if len(return_types) > 1:
retty = Tuple(return_types)
else:
retty = return_types[0]
argtypes = [arg.ty for arg in args if not arg.is_ptr]
return signature(retty, *argtypes)
from numba import njit, int64, float64
from numba.typed import List as L
from numba.types import Tuple, ListType as LT
import numpy as np
@njit(Tuple((int64[:,::1],LT(LT(int64)),LT(LT(int64))))(int64[:,::1], int64, int64), cache=True)
def compute_surface_mesh_adjs(edges, num_vertices, edges_per_face):
num_faces = edges.shape[0]//edges_per_face
adjs = np.zeros((num_faces, edges_per_face), dtype=np.int64)-1
vtx2vtx = L()
vtx2face = L()
for k in range(num_vertices):
tmp1 = L()
tmp2 = L()
tmp1.append(-1)
tmp2.append(-1)
vtx2vtx.append(tmp1)
vtx2face.append(tmp2)
tmp = np.arange(num_faces)
faces_idx = np.repeat(tmp, edges_per_face)
map_ = dict()
support_set = set()
map_[(-1,-1)] = -1
support_set.add((-1,-1))
for i in range(edges.shape[0]):
"Multicopter dynamics" import numpy as np import numba as nb from .. import quat from ..utis import env_param, polyval, cross # Common type definitions from numba.types import Tuple, none StateT = nb.float64[::1] ActionT = nb.float64[::1] TimeDeltaT = nb.float64 StateArrayT = nb.float64[:, ::1] ActionArrayT = nb.float64[:, ::1] DerivativesT = Tuple((StateArrayT, ActionArrayT)) eps = 1e-9 radius = 0.2 # m mass = 1.3 # kg num_rotors = 4 rpm_max = 20000 rotor_rpm_coeff = 6e1 friction_force = 1e-2 friction_torque = 1e-5 gravity = np.r_[0.0, 0.0, 9.82] # I = mk^2 where k is the radius of gyration inertia = mass * (0.275)**2 arm_length = radius rotor_angles = 2 * np.pi * (np.r_[0:num_rotors] - 1 / 2) / num_rotors
samples in each label class
Returns
-------
output : float
Gini impurity criterion in the node
"""
y_sum_sq = 0.0
w_samples_sq = w_samples * w_samples
for k in range(n_classes):
y_sum_sq += y_sum[k] * y_sum[k]
return 1.0 - y_sum_sq / w_samples_sq
@jit(
Tuple((float32, float32))(float32, float32, float32, float32[::1], float32[::1]),
nopython=NOPYTHON,
nogil=NOGIL,
boundscheck=BOUNDSCHECK,
fastmath=FASTMATH,
locals={
"y_sum_left_sq": float32,
"y_sum_right_sq": float32,
"w_samples_left_sq": float32,
"w_samples_right_sq": float32,
"gini_left": float32,
"gini_right": float32,
},
)
def gini_childs(n_classes, w_samples_left, w_samples_right, y_sum_left, y_sum_right):
"""Computes the gini impurity criterion in both left and right child nodes of a
from pandas.core.arrays import ExtensionArray, ExtensionOpsMixin
from collections import Iterable
from pandas.api.extensions import ExtensionDtype, take
from pandas.compat import set_function_name
from pandas.core.dtypes.generic import ABCSeries, ABCIndexClass
from pandas.core import ops
from pandas.core.dtypes.dtypes import registry
from pandas.core.dtypes.common import is_list_like, is_integer_dtype
import sparse
import operator
import numpy as np
from numba import jit
from numba.types import float64, Tuple, int64, void
@jit(Tuple((int64[:, :], float64[:], int64))(int64[:], float64[:], int64,
int64, float64),
parallel=True,
nogil=True,
nopython=True)
#@jit(parallel=True, nogil=True, nopython=True)
def _setitem(coords, data, shape, key, value):
# TODO: This seems like a very slow way to do it
data_iter = 0
key_copied = False
if value == 0.0:
return np.expand_dims(coords, 0), data, shape
newlen = data.shape[0] + 1
new_coords = np.empty((1, newlen), dtype=np.int64)
new_data = np.empty(newlen)
for i in range(newlen):
if data_iter >= 0:
# Compute the best bin and gain proxy obtained for the feature
find_best_split_along_feature(tree_context, node_context, feature, f,
candidate_split)
# If we found a candidate split along the feature
if candidate_split.found_split:
# And if it's better than the current one
if candidate_split.gain_proxy >= best_gain_proxy:
# Then we replace the best current split
copy_split(candidate_split, best_split)
best_gain_proxy = candidate_split.gain_proxy
f += 1
@jit(
[
Tuple((uintp, uintp))(TreeClassifierContextType, SplitClassifierType,
uintp, uintp, uintp, uintp),
Tuple((uintp, uintp))(TreeRegressorContextType, SplitRegressorType,
uintp, uintp, uintp, uintp),
],
nopython=True,
nogil=True,
locals={
"feature": uintp,
"bin_threshold": uint8,
"Xf": uint8[:],
"left_buffer": uintp[::1],
"right_buffer": uintp[::1],
"partition_train": uintp[::1],
"partition_valid": uintp[::1],
"n_samples_train_left": uintp,
"n_samples_train_right": uintp,
from cre.utils import _struct_from_meminfo, _meminfo_from_struct, _cast_structref, cast_structref, decode_idrec, lower_getattr, _struct_from_pointer, lower_setattr, lower_getattr, _pointer_from_struct
from cre.caching import gen_import_str, unique_hash, import_from_cached, source_to_cache, source_in_cache
from cre.condition_node import Conditions, ConditionsType, initialize_conditions
from cre.var import *
from cre.predicate_node import GenericAlphaPredicateNodeType, GenericBetaPredicateNodeType
from operator import itemgetter
from copy import copy
i8_arr = i8[::1]
i8_arr_2d = i8[:, ::1]
list_i8_arr = ListType(i8_arr)
# list_i8_arr_2d = ListType(optional(i8_arr_2d))
i8_i8_arr_tuple = Tuple((i8, i8[:, ::1]))
list_i8_i8_arr_tuple = ListType(i8_i8_arr_tuple)
i8_x2 = UniTuple(i8, 2)
@njit(cache=True)
def filter_alpha(term, inds):
png = cast_structref(GenericAlphaPredicateNodeType, term.pred_node)
mi = _meminfo_from_struct(png)
return png.filter_func(mi, term.link_data, inds, term.negated)
@njit(cache=True)
def filter_beta(term, l_inds, r_inds):
png = cast_structref(GenericBetaPredicateNodeType, term.pred_node)
r"""Distance functions for potentials.
Some of the functions also compute rotational moments for computing torque like
.. math::
\mathbf{M} = \mathbf{r}_{\mathrm{moment}} \times (\mathbf{f}_{}^{soc} + \mathbf{f}_{}^{c})
"""
import numba
import numpy as np
from numba import float64
from numba.types import Tuple, UniTuple
from crowddynamics.core.vector2D import length, rotate90, dot
@numba.jit([Tuple((float64, float64[:]))(float64[:], float64,
float64[:], float64)],
nopython=True, nogil=True, cache=True)
def distance_circles(x0, r0, x1, r1):
r"""
Skin-to-Skin distance :math:`h` with normal :math:`\mathbf{\hat{n}}`
between two circles.
.. math::
h &= \|\mathbf{x}_0 - \mathbf{x}_1\| - (r_0 + r_1) \\
\mathbf{\hat{n}} &= \frac{\mathbf{x}_0 - \mathbf{x}_1}{\|\mathbf{x}_0 - \mathbf{x}_1\|}
Args:
x0 (numpy.ndarray):
r0 (float):
x1 (numpy.ndarray):
count = tree.nodes.counts[idx_node, c]
return n_samples == count
@njit(uint32(TreeClassifier.class_type.instance_type, uint32, float32[::1]))
def node_get_child(tree, idx_node, x_t):
feature = tree.nodes.feature[idx_node]
threshold = tree.nodes.threshold[idx_node]
if x_t[feature] <= threshold:
return tree.nodes.left[idx_node]
else:
return tree.nodes.right[idx_node]
@njit(
Tuple((float32, float32))(TreeClassifier.class_type.instance_type, uint32,
uint32))
def node_range(tree, idx_node, j):
# TODO: do the version without memory...
if tree.nodes.n_samples[idx_node] == 0:
raise RuntimeError("Node has no range since it has no samples")
else:
return (
tree.nodes.memory_range_min[idx_node, j],
tree.nodes.memory_range_max[idx_node, j],
)
@njit(
float32(TreeClassifier.class_type.instance_type, uint32, float32[::1],
float32[::1]))
def node_compute_range_extension(tree, idx_node, x_t, extensions):
# Using loops since the numpy functions such as `numpy.isin` and `numpy.intersect1d` are not supported by numba
s_xy = 0.
for i in x:
for j in y:
if i == j:
s_xy += 1.
break
cs = s_xy / ((s_x * s_y) ** 0.5)
# Clip values to the range `[-1, 1]`, the domain of arc-cosine
dist = np.arccos(max(-1., min(1., cs)))
return dist
@njit(Tuple((int64[:, :], float64[:, :]))(int64[:, :], float64[:, :]))
def remove_self_neighbors(index_neighbors_, distance_neighbors_):
"""
Given the index and distances of k nearest neighbors of a list of query points, remove points from their
own neighbor list.
:param index_neighbors_: numpy array of the index of `k` neighbors for a list of points. Has shape `(n, k)`,
where `n` is the number of query points.
:param distance_neighbors_: numpy array of the distance of `k` neighbors for a list of points.
Also has shape `(n, k)`.
:return: (index_neighbors, distance_neighbors), where each of them has shape `(n, k - 1)`.
"""
n, k = index_neighbors_.shape
index_neighbors = np.zeros((n, k - 1), dtype=index_neighbors_.dtype)
distance_neighbors = np.zeros((n, k - 1), dtype=distance_neighbors_.dtype)
"""
x_0 = np.roll(img, -1, axis=1).T
x_1 = np.roll(img, 1, axis=1).T
y_0 = np.roll(img, -1, axis=0).T
y_1 = np.roll(img, 1, axis=0).T
# we do a lot of transposing before and after here because sums in the
# energy function happen along the first dimension by default when we
# want them to be happening along the last (summing the colors)
en_map = sum(pow((x_0 - x_1), 2) + pow((y_0 - y_1), 2)).T
return en_map
@njit(Tuple((Array(int64, 2, 'A'), int64))(Array(uint8, 2, 'A')))
def cumulative_energy(energy):
"""
https://en.wikipedia.org/wiki/Seam_carving#Dynamic_programming
Parameters
==========
energy: 2-D numpy.array(uint8)
Produced by energy_map
Returns
=======
tuple of 2 2-D numpy.array(int64) with shape (height, width).
paths has the x-offset of the previous seam element for each pixel.
path_energies has the cumulative energy at each pixel.
"""
Since numba's jitclass support for class inheritance is inexistant, we use this
function both in the constructors of SplitClassifier and SplitRegressor.
Parameters
----------
split : SplitClassifier or SplitRegressor
The split to be initialized by this function
"""
split.bins = np.empty(max_n_bins, dtype=np.uint64)
split.bin_partition = np.empty(max_n_bins, dtype=np.uint64)
reset_split(split)
@jit(
Tuple((float32, float32))(
uint64, float32, float32, float32, float32[::1], float32[::1]
),
nopython=NOPYTHON,
nogil=NOGIL,
boundscheck=BOUNDSCHECK,
fastmath=FASTMATH,
locals={},
)
def apply_childs_impurity_function_clf(
criterion,
n_classes,
w_samples_train_left,
w_samples_train_right,
y_sum_left,
y_sum_right,
):
for each_node in top_layer:
for e in graph.indices[graph.indptr[each_node]:graph.indptr[each_node + 1]]:
directConnectionScore[e] +=1
directConnectionScore = np.log(directConnectionScore+2)
# get node attribute score
beta = 0.25
node_score = beta * np.log10(np.max(graph.node_attr)/graph.node_attr)
contentscore = node_score + directConnectionScore
return contentscore
@njit(Tuple((int32, float32))(DecSparseGraphSpec(), int32[:]), nogil=True)
def vertex_nomination__kernel__(G, seeds):
# context score : computed via distance map from multi BFS
bfs_results = multi_bfs(G, seeds)
context_sim = 1/bfs_results
# content score : placeholder computed via a random number generator
content_sim = getContentScore(G, seeds)
# fusion score : content_sim * context_sim
fusion_score = np.multiply(content_sim, context_sim)
# remove original seeds from ranking
for seed in seeds:
fusion_score[seed] = np.float64(-1)
Parameters
----------
records : Records
A records dataclass containing the stack of node records
Returns
-------
output : bool
Returns True if there are remaining records in the stack, False otherwise
"""
return records.top > 0
@jit(
Tuple((intp, uintp, boolean, float32, uintp, uintp, uintp, uintp))(RecordsType),
nopython=NOPYTHON,
nogil=NOGIL,
boundscheck=BOUNDSCHECK,
fastmath=FASTMATH,
locals={"stack_top": record_type},
)
def pop_node_record(records):
"""Pops (removes and returns) a node record from the stack of records.
Parameters
----------
records : Records
A records dataclass containing the stack of node records
Returns
s_len = 2**(l_max + 1)
s = np.zeros(c.shape[:-1] + (s_len, ), dtype=c.dtype)
pos_sums = np.arange(l_max, dtype=np.int64)
pos_sums[:] = 2**(l_max - pos_sums)
pos_sums = np.cumsum(pos_sums)
gamma = np.zeros_like(c)
return (x, y, ix, iy, vx, vy, unbiased, iy_reord, c, sx_c, sy_c, c_sum,
l_max, s, pos_sums, gamma)
_get_impl_args_compiled = numba.njit(Tuple(
(input_array, input_array, int64[:], int64[:], float64[:], float64[:],
boolean, int64[:], float64[:, :], float64[:, :], float64[:, :],
float64[:], int64, float64[:, :], int64[:], float64[:, :]))(input_array,
input_array,
boolean),
cache=True)(_get_impl_args)
impls_dict = {
CompileMode.AUTO:
((_get_impl_args_compiled, _distance_covariance_sqr_avl_impl_compiled),
(_get_impl_args, _distance_covariance_sqr_avl_impl)),
CompileMode.NO_COMPILE:
((_get_impl_args, _distance_covariance_sqr_avl_impl), ),
CompileMode.COMPILE_CPU:
((_get_impl_args_compiled, _distance_covariance_sqr_avl_impl_compiled), )
}
Tuple((float64[:, :, :], int64, int64))(
# D N W X
int64,
int64,
float64[:, :],
float64[:, :],
# C R kest Z
uint8[:],
int64[:],
int64,
float64[:, :, :],
# Breal Bint Bpos Bcat Bord
float64[:, :, :],
float64[:, :, :],
float64[:, :, :],
float64[:, :, :, :],
float64[:, :, :],
# Bcount u wint theta thetal thetah
float64[:, :, :],
float64[:, :, :],
int64,
float64[:, :],
float64[:],
float64[:],
# maxX meanX s2y s2u syd n_samples
float64[:],
float64[:],
float64,
float64,
float64,
int64),
import numpy as np
from numba import jit
from numba.types import void, Tuple, int32, boolean, float32
from . import utils
from .planar_graph import PlanarGraph, planar_graph_nb_type
@jit(Tuple((boolean[:], boolean[:], int32[:]))(planar_graph_nb_type, int32[:],
int32),
nopython=True)
def get_tree_cycle_masks_and_next_edge_indices_in_path_to_cycle(
graph, parent_edge_indices, non_tree_edge_index):
cycle_vertices_mask = utils.repeat_bool(False, graph.size)
cycle_edges_mask = utils.repeat_bool(False, graph.edges_count)
next_edge_indices_in_path_to_cycle = utils.repeat_int(-1, graph.size)
non_tree_edge_vertex1 = graph.edges.vertex1[non_tree_edge_index]
non_tree_edge_vertex2 = graph.edges.vertex2[non_tree_edge_index]
cycle_vertices_mask[non_tree_edge_vertex1] = True
cycle_vertices_mask[non_tree_edge_vertex2] = True
no_vertices_overlapped_yet = True
for start_vertex in [non_tree_edge_vertex1, non_tree_edge_vertex2]:
current_vertex = start_vertex
current_edge_index = parent_edge_indices[current_vertex]
chunk = 1
elif n_cols % num_threads > 0:
chunk = (n_cols // num_threads) + 1
else:
chunk = (n_cols // num_threads)
for i in range(n_rows):
for j in range((tid) * chunk, min(chunk * (tid + 1), n_cols)):
tmp = 0.0
for k in range(A.indptr[j], A.indptr[j + 1]):
tmp += B[i, A.indices[k]] * A.data[k]
C[i, j] = tmp
#C[:,i] = np.dot(,A.data[A.indptr[i]:A.indptr[i+1]])
@njit(Tuple((int64[:], int64[:], int64[:]))(int64[:], int64[:]),
nogil=True,
pipeline_class=DEC_Pipeline)
def list_intersection(list1, list2):
len1 = len(list1)
len2 = len(list2)
x = np.zeros(max(len1, len2), dtype=np.int64)
xidx = np.zeros(max(len1, len2), dtype=np.int64)
xjdx = np.zeros(max(len1, len2), dtype=np.int64)
i, j, lenx = 0, 0, 0
while (i < len1) and (j < len2):
if list1[i] == list2[j]:
x[lenx] = list1[i]
xidx[lenx] = i
xjdx[lenx] = j
#COPIED FROM NUMBERT EXPERIMENTAL BRANCH
from numba import types, njit, u1, u2, u4, u8, i8, i2, literally
from numba.types import Tuple, void
from numba.experimental.structref import _Utils, imputils
from numba.extending import intrinsic
from numba.core import cgutils
from llvmlite.ir import types as ll_types
#### idrec encoding ####
@njit(Tuple([u2, u8, u1])(u8), cache=True)
def decode_idrec(idrec):
t_id = idrec >> 48
f_id = (idrec >> 8) & 0x000FFFFF
a_id = idrec & 0xF
return (t_id, f_id, a_id)
@njit(u8(u2, u8, u1), cache=True)
def encode_idrec(t_id, f_id, a_id):
return (t_id << 48) | (f_id << 8) | a_id
meminfo_type = types.MemInfoPointer(types.voidptr)
@intrinsic
def lower_setattr(typingctx, inst_type, attr_type, val_type):
if (isinstance(attr_type, types.Literal)
# If we found a candidate split along the feature
if candidate_split.found_split:
# And if it's better than the current one
if candidate_split.gain_proxy >= best_gain_proxy:
# Then we replace the best current split
copy_split(candidate_split, best_split)
best_gain_proxy = candidate_split.gain_proxy
f += 1
# TODO: Compute the true information gain and save it somewhere ? But it's only
# useful for root ?
return best_split
@jit(
Tuple((uintp, uintp))(TreeContextType, SplitType, uintp, uintp, uintp, uintp),
nopython=True,
nogil=True,
locals={
"feature": uintp,
"bin_threshold": uint8,
"Xf": uint8[::1],
"left_buffer": uintp[::1],
"right_buffer": uintp[::1],
"partition_train": uintp[::1],
"partition_valid": uintp[::1],
"n_samples_train_left": uintp,
"n_samples_train_right": uintp,
"n_samples_valid_left": uintp,
"n_samples_valid_right": uintp,
"pos_train": uintp,
import numpy as np
import sys, math
import multiprocessing
from numba import float64, float32, jit, boolean, vectorize, int64
from numba.types import UniTuple, Tuple, void
from bimodal_mixture_model import make_bimodal_params_numba, value_to_bimodal_prob_numba
from cluster_init import assign_cells_init, spectral_cluster_cosine, ward_cluster_cosine, spectral_cluster, boolean_spectral_cluster, nmf_init, kmeans_cluster
from likelihood import full_log_likelihood_single, greedy_optimize_i_numba_single, greedy_optimize_j_numba_single, sgn
import random
import pickle
from sklearn import model_selection
@jit(Tuple((float64[:], float64[:, :]))(float64[:, :], float64[:, :]),
nopython=True)
def get_memo_params(
i_cells: np.ndarray,
j_genes: np.ndarray,
):
z_count = np.dot(i_cells, j_genes.T)
z = sgn(z_count)
z_arr = np.sum(z, 0)
return z_arr, z_count
def estimate_epi_data(i_cells: np.ndarray, j_genes: np.ndarray):
z = np.dot(i_cells, j_genes.T).astype(bool).astype(float)
z_total = np.sum(z, 0) * (1.0 / (i_cells.shape[0]))
return z_total
@generated_jit(cache=True)
@overload(operator.eq)
def var_eq(left_var, right_var):
return comparator_helper("==", left_var, right_var)
@generated_jit(cache=True)
@overload(operator.ne)
def var_ne(left_var, right_var):
return comparator_helper("==", left_var, right_var, True)
## dnf_type ##
pterm_list_type = ListType(PTermType)
ab_conjunct_type = Tuple((pterm_list_type, pterm_list_type))
dnf_type = ListType(ab_conjunct_type)
## distr_dnf_type ##e
pterm_list_list_type = ListType(pterm_list_type)
distr_ab_conjunct_type = Tuple(
(pterm_list_list_type, pterm_list_list_type, i8[:, :]))
distr_dnf_type = ListType(distr_ab_conjunct_type)
conditions_fields_dict = {
### Fields that are filled on in instantiation ###
# The variables used by the condition
'vars': ListType(GenericVarType),
# The Disjunctive Normal Form of the condition but organized
import numpy as np
import scipy.stats as st
import matplotlib
import matplotlib.pyplot as plt
import time
from numba import jit, void, f8, i8
from numba.types import UniTuple, Tuple
import datetime
from pathlib import Path
"""
Time step functions
"""
# calc dt for host dynamics, creating integer number of steps within dt of bacterial dynamics
@jit(Tuple((f8, i8))(f8, f8[::1], f8), nopython=True)
def calc_dynamic_timestep(hostProp, dtVec, dtBac):
# calculate smallest time step ever needed
dt = mlsg.calc_max_time_step(hostProp, dtVec)
if dt == dtVec.min():
print("warning: min dt reached")
# make sure that there is integer number of time steps
numSubStep = int(np.ceil(dtBac / dt))
dt = dtBac/numSubStep
return (dt, numSubStep)
"""
Init functions
"""
#initialize community
def init_comm(model_par):
if data_test is None:
labels_pred = knn_model.predict(data, is_train=True)
# error rate
mask = labels_pred != labels
err_rate = float(mask[mask].shape[0]) / labels.shape[0]
else:
labels_pred = knn_model.predict(data_test, is_train=False)
# error rate
mask = labels_pred != labels_test
err_rate = float(mask[mask].shape[0]) / labels_test.shape[0]
return err_rate
@njit(Tuple((float64[:], float64[:, :]))(int64[:, :], int64[:], int64), fastmath=True)
def neighbors_label_counts(index_neighbors, labels_train, n_classes):
"""
Given the index of neighboring samples from the training set and the labels of the training samples,
find the label counts among the k neighbors and assign the label corresponding to the highest count as the
prediction.
:param index_neighbors: numpy array of shape `(n, k)` with the index of `k` neighbors of `n` samples.
:param labels_train: numpy array of shape `(m, )` with the class labels of the `m` training samples.
:param n_classes: (int) number of distinct classes.
:return:
- labels_pred: numpy array of shape `(n, )` with the predicted labels of the `n` samples. Needs to converted
to type `np.int` at the calling function. Numba is not very flexible.
- counts: numpy array of shape `(n, n_classes)` with the count of each class among the `k` neighbors.
"""
from numba import njit, int64, float64
from numba.typed import List as L
from numba.types import Tuple, List, ListType as LT
import numpy as np
#edges, vtx2vtx, vtx2edge, vtx2poly, edge2vtx, edge2edge, edge2poly, poly2vtx, poly2edge, poly2poly , LT(LT(int64)),LT(LT(int64)), LT(LT(int64)), int64[:,::1], LT(LT(int64)), LT(LT(int64)), int64[:,::1], int64[:,::1], int64[:,::1]
@njit(Tuple(
(int64[:, ::1], LT(LT(int64)), LT(LT(int64)), LT(LT(int64)), int64[:, ::1],
LT(LT(int64)), LT(LT(int64)), int64[:, ::1], int64[:, ::1],
int64[:, ::1]))(int64, int64[:, ::1]),
cache=True)
def get_connectivity_info_surface(num_vertices, polys):
vtx2vtx = L()
vtx2edge = L()
vtx2poly = L()
edge2edge = L()
edge2poly = L()
for i in range(num_vertices):
tmp1 = L()
tmp2 = L()
tmp3 = L()
tmp1.append(-1)
tmp2.append(-1)
tmp3.append(-1)
vtx2vtx.append(tmp1)
vtx2edge.append(tmp2)
vtx2poly.append(tmp3)
tree_edges_mask[incident_edge_index] = True
_construct_tree_edges_mask = utils.make_traverse_graph_via_bfs(_add_edge_to_tree, boolean[:])
@jit(boolean[:](int32, planar_graph_nb_type), nopython=True)
def construct_bfs_tree_edges_mask(root, graph):
tree_edges_mask = utils.repeat_bool(False, graph.edges_count)
used_vertex_flags = utils.repeat_bool(False, graph.size)
_construct_tree_edges_mask(root, graph, used_vertex_flags, tree_edges_mask)
return tree_edges_mask
@jit(Tuple((int32[:], int32[:]))(planar_graph_nb_type, int32[:], int32, boolean[:]), nopython=True)
def _get_ordered_bfs_subtree_adjacencies_and_incidence_indices(graph, bfs_levels,
subtree_max_level, bfs_tree_edges_mask):
bfs_subtree_edges_mask = bfs_tree_edges_mask
for edge_index in range(graph.edges_count):
edge_vertex1 = graph.edges.vertex1[edge_index]
edge_vertex2 = graph.edges.vertex2[edge_index]
if max(bfs_levels[edge_vertex1], bfs_levels[edge_vertex2]) > subtree_max_level:
bfs_subtree_edges_mask[edge_index] = False
ordered_bfs_subtree_adjacencies_list = []
ordered_bfs_subtree_incidence_indices_list = []
import numpy as np
from numba import njit, float64, int64
from numba.types import Tuple
@njit(Tuple((float64[:,::1], int64[:,::1]))(float64[:,::1],int64[:,::1]), cache=True)
def remove_duplicated_vertices(vertices, faces):
vtx_dictionary = dict()
support_set = set()
vtx_dictionary[(-1.,-1.,-1.)] = -1.
support_set.add((-1.,-1.,-1.))
new_vertices = np.zeros(vertices.shape, dtype=np.float64)
j=0
for i in range(vertices.shape[0]):
v = (vertices[i][0], vertices[i][1], vertices[i][2])
if v not in support_set:
vtx_dictionary[v] = i
support_set.add(v)
new_vertices[j] = vertices[i]
j+=1
else:
idx = vtx_dictionary[v]
r = np.where(faces==i)
for k in zip(r[0], r[1]):
faces[k[0]][k[1]] = idx
import numpy as np
from numba import njit, float64, int64
from numba.types import Tuple
@njit(Tuple((float64[:, ::1], int64[:, ::1]))(float64[:, ::1], int64[:, ::1]),
cache=True)
def remove_duplicated_vertices(vertices, faces):
vtx_dictionary = dict()
support_set = set()
vtx_dictionary[(-1., -1., -1.)] = -1.
support_set.add((-1., -1., -1.))
new_vertices = np.zeros(vertices.shape, dtype=np.float64)
j = 0
for i in range(vertices.shape[0]):
v = (vertices[i][0], vertices[i][1], vertices[i][2])
if v not in support_set:
vtx_dictionary[v] = i
support_set.add(v)
new_vertices[j] = vertices[i]
j += 1
else:
idx = vtx_dictionary[v]
r = np.where(faces == i)
for k in zip(r[0], r[1]):
