Source code for ddg.nets._utils

import itertools
from collections.abc import Iterable
from functools import partial

import numpy as np

import ddg
import ddg.nets._conversion as netsconv
from ddg.math import euclidean
from ddg.nets._decorators import applicable_to_netcollection, domain_util, net_util
from ddg.nets._domain import (
    DiscreteDomain,
    DiscreteRectangularDomain,
    SmoothRectangularDomain,
    _is_bounded_interval,
)
from ddg.nets._net import (
    DiscreteCurve,
    DiscreteNet,
    EmptyNet,
    Net,
    NetCollection,
    PointNet,
    SmoothCurve,
    SmoothNet,
)


[docs]def compose(f, n): """The composition x -> f(n(x)) adapted to nets and net collections. Parameters ---------- f : callable or numpy.ndarray n : Net or NetCollection Returns ------- Net or NetCollection The same type as `n`. Examples -------- >>> import numpy as np >>> from ddg.nets import SmoothNet >>> from ddg.nets import compose >>> line = SmoothNet(lambda t: np.array([t, 0]), [(0, np.pi, True)]) >>> line(2) array([2, 0]) >>> translated_line = compose(lambda x: x + (0, 1), line) >>> translated_line(2) array([2, 1]) >>> linearly_transformed_line = compose(np.array([[2, 1], [0, 1]]), line) >>> linearly_transformed_line(2) array([4, 0]) Unlike .. code-block:: python def f_after_n(x): return f(n(x)) ``compose`` retains the net domain(s) by returning a net or net collection. >>> type(linearly_transformed_line) <class 'ddg.nets._net.SmoothCurve'> >>> linearly_transformed_line.domain SmoothInterval([0.0, 3.14..., True]) The parameter `n` is required to be a net or a net collection. If it isn't, then there is no need to retain a domain, so .. code-block:: python def f_after_n(x): return f(n(x)) suffices. See Also -------- ddg.nets.embed ddg.nets.homogenize ddg.nets.dehomogenize """ if isinstance(f, np.ndarray): def g(*args, **kwargs): return f @ n.fct(*args, **kwargs) elif callable(f): def g(*args, **kwargs): return f(n.fct(*args, **kwargs)) else: raise TypeError("f must be callable or an instance of numpy.ndarray") # NetCollection isn't actually a Net. # EmptyNet and PointNet have constructors that require # points/coordinates/values rather than functions. if isinstance(n, NetCollection): return NetCollection([compose(f, subnet) for subnet in n], n.name) elif isinstance(n, EmptyNet): return EmptyNet(g(), n.name) elif isinstance(n, PointNet): return PointNet(g(), n.name) elif isinstance(n, Net): return type(n)(g, n.domain, name=n.name) else: raise TypeError(f"n must be an instance of {Net}")
[docs]def homogenize(net): """ Homogenize the values of given net. Parameters ---------- net : ddg.nets.Net The net to operate on. Returns ------- A net of the same type. See Also -------- Net Notes ----- * It is assumed that the values of the net are of type `np.array`. """ return compose(ddg.math.projective.homogenize, net)
[docs]def dehomogenize(net): """ Dehomogenize the values of given net. Parameters ---------- net : ddg.nets.Net The net to operate on. Returns ------- A net of the same type. See Also -------- Net Notes ----- * It is assumed that the values of the net are of type `np.array`. """ return compose(ddg.math.projective.dehomogenize, net)
[docs]def embed(net, level=0, component=-1): """ Embeds the given net in a space one dimension higher than before. Parameters ---------- net : ddg.nets.Net The net to operate on. level : float (default 0) component : int (default=-1) Component to add Returns ------- A net of the same type. See Also -------- Net Notes ----- * It is assumed that the values of the net are of type `np.array`. """ return compose(partial(euclidean.embed, index=component, level=level), net)
[docs]def vertices(net): """ Get all vertex values from a given net or net collection. Parameters ---------- net : ddg.nets.DiscreteNet or NetCollection The net or net collection to obtain all the vertices from Returns ------- Generator Generator containing all vertex values. """ # Check whether domain of net is bounded if isinstance(net, DiscreteNet) and not net.domain.bounded: raise TypeError("Domain of net is unbounded.") elif isinstance(net, NetCollection): for n in net: if isinstance(n, DiscreteNet) and not n.domain.bounded: raise TypeError("Domain of net is unbounded.") if ( isinstance(net, DiscreteNet) or isinstance(net, PointNet) or isinstance(net, EmptyNet) ): return iter(net) elif isinstance(net, NetCollection): return itertools.chain(*map(vertices, net)) else: raise TypeError("Only supports DiscretNet and NetCollection.")
[docs]@net_util def surface_of_revolution(curve, axis=2, name="Surface of Revolution"): """ Generate a surface of revolution from a planar curve Parameters ---------- curve : ddg.nets.SmoothNet Planar curve. axis : 0, 1, 2 Axis of rotation. name: str Name of the surface. Returns ------- ddg.nets.SmoothNet Surface of revolution. """ def fct(u, v): x = np.array([curve(u)[0] * np.cos(v), curve(u)[0] * np.sin(v)]) if axis == 2: x = np.append(x, curve(u)[1]) else: x = np.insert(x, axis, curve(u)[1]) return x # Might need a change when Intervals are removed intervals = [curve.domain.recover()] intervals.append([0, 2 * np.pi, True]) return (fct, intervals)
[docs]@net_util def cylinder(curve, axis=2, name="Cylinder"): """ Generate a cylinder from a planar base curve Parameters ---------- curve : ddg.nets.SmoothNet Planar curve. axis : 0, 1, 2 Axis of the cylinder name: str Name of the surface. Returns ------- ddg.nets.SmoothNet Cylinder. """ def fct(u, v): if axis == 2: x = np.append(curve(u), v) else: x = np.insert(curve(u), axis, v) return x # Might need a change when Intervals are removed intervals = [curve.domain.recover()] intervals.append([-np.inf, np.inf]) return (fct, intervals)
[docs]@net_util def cone(curve, cone_vertex, name="Cone"): """ Generate a cone over a given cone. Parameters ---------- curve : ddg.nets.SmoothNet Spatial curve. cone_vertex : numpy.array Coordinates of a point in space name: str Name of the surface. Returns ------- ddg.nets.SmoothNet Cone Notes ----- * The cone is generated by connecting all the pointes of the curve with the cone vertex. """ def fct(u, v): x = np.multiply(curve(u), v) + np.multiply(cone_vertex, 1 - v) return x intervals = [curve.domain.recover()] intervals.append([-np.inf, np.inf]) return (fct, intervals)
[docs]def coordinate_hypersurface(net, component, value=None): """Generate the (parameterized) level set of a coordinate-direction. Parameters ---------- net : SmoothNet or DiscreteNet with rectangular domain Source net. component : numpy.ndarray Coordinate to set to a constant value. value : Value for the chosen component of the domain. If no value is given the output net will depend on the value as a Parameter. Returns ------- ddg.nets.Net The hypersurface as a net. Notes ----- * The hypersurface is generated by setting one coordinates to a constant given value. """ if value is None: default_value = (net.domain[component][0] + net.domain[component][1]) / 2 intervals = ( net.domain.recover()[:component] + net.domain.recover()[component + 1 :] ) domain = net.domain.__class__(intervals) def fct(*args, value=0.0, **kwargs): args = args[:component] + (value,) + args[component:] return net(*args, **kwargs) if isinstance(net.domain, SmoothRectangularDomain): new_net = SmoothNet(fct, domain) elif isinstance(net.domain, DiscreteRectangularDomain): new_net = DiscreteNet(fct, domain) else: intervals = ( net.domain.recover()[:component] + net.domain.recover()[component + 1 :] ) domain = net.domain.__class__(intervals) def fct(*args, **kwargs): args = args[:component] + (value,) + args[component:] return net(*args, **kwargs) if isinstance(net.domain, SmoothRectangularDomain): new_net = SmoothNet(fct, domain) elif isinstance(net.domain, DiscreteRectangularDomain): new_net = DiscreteNet(fct, domain) return new_net
[docs]def coordinate_line(net, point, direction): """Gives a curve on the net starting at a point in a given direction. Parameters ---------- net : SmoothNet or DiscreteNet with rectangular domain Given net. point : Iterable of Real Point in the domain from where to start the curve. direction : numpy.array, or int Direction in the domain as an array, or component as int. Returns ------- ddg.nets.Net A line on the net. Notes ----- * If direction is given as an int, the periodicity of the net in the given direction is inherited by the curve. """ point = np.array(point, dtype=np.float64) if isinstance(direction, int): direction = np.eye(net.dimension)[direction] direction = np.array(direction, dtype=np.float64) periodic = False if isinstance(direction, int): periodic = direction in net.domain.periodicity bounds = [] for v, (a, b), co in zip(direction, net.domain.intervals, point): if v != 0: if v < 0: a, b = b, a bounds.append([(a - co) / v, (b - co) / v]) bounds = np.transpose(bounds) intervals = [[np.nanmax(bounds[0]), np.nanmin(bounds[1]), periodic]] domain = net.domain.__class__(intervals) fct = lambda s: net(*(point + (s * direction))) if isinstance(net, SmoothNet): coordinate_line = SmoothCurve(fct, domain) elif isinstance(net, DiscreteNet): coordinate_line = DiscreteCurve(fct, domain) else: raise ValueError("Only works for smooth or discrete nets.") return coordinate_line
[docs]def coordinate_lines(net, direction, sampling, anchor=None, atol=None): """Get coordinate lines in a given coordinate direction. Parameters ---------- net : ddg.nets.SmoothNet Given net (must be SmoothNet with rectangular domain). direction : int Direction of the coordinate lines (component number). sampling : list, int or float See sample_smooth_domain anchor : list, tuple or NONE Point inside the domain of the net. The sampling will be chosen such that the point is part of it. atol : float or None (default=None) Is used as tolerance for symmetric sampling. See: sample_interval This function uses the global tolerance defaults if `atol` or `rtol` are set to None. See :py:mod:`ddg.nonexact` for details. Returns ------- ddg.nets.NetCollection Coordinate lines as a NetCollection. Notes ----- * They will be stepsize apart in all other directions. * A set anchor precedes symmetric sampling """ if not isinstance(direction, int): raise ValueError("Direcion should be given by one int value.") anchor = np.array(anchor) dom = ( net.domain.recover() ) # list of intervals with periodictiy information, len(dom)= dimension direction_value = dom[direction][ 0 ] # one point of the domain, we need it for embedding small_domain = delete_direction(net.domain, direction) if anchor is not None: anchor = np.delete(anchor, direction) sampled_domain = netsconv.sample_smooth_domain( small_domain, sampling, anchor=anchor, atol=atol ) sampled_domain = embed(sampled_domain, level=direction_value, component=direction) nc = NetCollection() for point in sampled_domain: cl = coordinate_line(net, point, direction) nc.add(cl) return nc
[docs]def coordinate_grid(net, sampling, anchor=None, atol=None): """Get a grid of your net of given sampling (fineness in each direction) Parameters ---------- net: ddg.nets.SmoothNet Given net (must be SmoothNet with RectangularDomain). sampling : list, int or float Determines how the domain is sampled. See sample_smooth_domain. anchor : numpy.ndarray, or None Point inside the domain of the net. The sampling will be chosen such that the point is part of it. atol : float or None (default=None) Is used as tolerance for symmetric sampling. See: sample_interval This function uses the global tolerance defaults if `atol` or `rtol` are set to None. See :py:mod:`ddg.nonexact` for details. Returns ------- list of NetCollections of coordinate lines in every direction """ nclist = [] dim = net.domain.dimension sampling_list = netsconv.sampling_decomposer(sampling, atol, anchor, dim) for i in range(dim): # this if else stuff is here because the sampling decomposer works a # little weird. sometimes the stepsize is in a list, sometimes not if isinstance(sampling_list[0][i], Iterable): coordlines = coordinate_lines( net, i, [sampling_list[0][i][0], sampling_list[1][i]], anchor=sampling_list[3], atol=sampling_list[2], ) # creates coordlines of net in direction i else: coordlines = coordinate_lines( net, i, [sampling_list[0][i], sampling_list[1][i]], anchor=sampling_list[3], atol=sampling_list[2], ) # creates coordlines of net in direction i nclist.append(coordlines) return nclist
[docs]def diagonal_lines(net, direction, sampling, anchor=None, atol=None): """Get coordinate lines in a given coordinate direction. Parameters ---------- net : ddg.nets.SmoothNet Given 2-dim net (must be SmoothNet with rectangular domain). Only works for bounded nets. direction: np.array of dim 2 vector consisting of 1 and -1, determining the direction of the diagonals sampling : list of length 2 of form [number, "option"] or float (stepsize) See ddg.nets.conversion.sample_smooth_domain anchor : numpy.ndarray, or None Point inside the domain of the net. The sampling will be chosen such that the point is part of it. atol : float or None (default=None) Is used as tolerance for symmetric sampling. See: sample_interval This function uses the global tolerance defaults if `atol` or `rtol` are set to None. See :py:mod:`ddg.nonexact` for details. Returns ------- ddg.nets.NetCollection Coordinate lines as a NetCollection. Notes ----- * They will be stepsize apart in all other directions. * A set anchor precedes symmetric sampling """ nc = NetCollection() dom = ( net.domain.recover() ) # list of intervals with periodictiy information, len(dom)= dimension if len(dom) != 2: raise ValueError("Only works for 2-dim Nets") # decoding our domain as [x1,x2]x[y1,y2] x1 = dom[0][0] x2 = dom[0][1] y1 = dom[1][0] y2 = dom[1][1] sample_left = False sample_right = False # 1st case: diagonals in direction of vector [1,1]. we need to sample the # left side of the rectangle if np.array_equal(direction, [1, 1]) or np.array_equal(direction, [-1, -1]): sample_left = True # 2nd case: diagonals in direction [1,-1]. we need to sample the right side # of the rectangle elif np.array_equal(direction, [1, -1]) or np.array_equal(direction, [-1, 1]): sample_right = True else: raise ValueError("Direction must consist of 1 and -1 values.") bottom_side = delete_direction(net.domain, 1) if sample_left: extended_bottom_side = modify_direction(bottom_side, 0, [x1 - (y2 - y1), x2]) if isinstance(sampling, Iterable): if len(sampling) != 2: raise ValueError( "You can only specify the sampling of one direction of your domain." " Only works for bounded domains." ) if anchor is not None: raise NotImplementedError( "Options 's' and 't' not compatible with anchor" ) sampled_extended_bottom = netsconv.sample_smooth_domain( extended_bottom_side, sampling, atol=atol ) # Now we need to add the coordinate lines at the points in the # first interval(bottom side), and the intersection points of the # diags with the second interval (left side) for idx in sampled_extended_bottom.domain.traverser: bottom_point = sampled_extended_bottom(idx[0])[0] if bottom_point < x1: # we want to find the points on the "left side" we want to # send coord lines from embedded_point = [x1, x1 - bottom_point + y1] cl = coordinate_line(net, embedded_point, direction) nc.add(cl) elif bottom_point >= x1: # we just need to embed the bottom points embedded_point = [bottom_point, y1] cl = coordinate_line(net, embedded_point, direction) nc.add(cl) elif isinstance(sampling, float) or isinstance(sampling, int): if isinstance(sampling, int): sampling = float(sampling) if anchor is None: bottom_anchor = x1 if anchor is not None: bottom_anchor = anchor[0] - (anchor[1] - y1) sampled_extended_bottom = netsconv.sample_smooth_domain( extended_bottom_side, sampling, anchor=[bottom_anchor], atol=atol ) # CALCULATING sampled points of LEFT side as intersection of the # diaglines in extended bottom side for idx in sampled_extended_bottom.domain.traverser: bottom_point = sampled_extended_bottom(idx[0])[0] if bottom_point < x1: # THEN WE WANT TO SAMPLE THE LEFT SIDE embedded_point = [x1, x1 - bottom_point + y1] cl = coordinate_line(net, embedded_point, direction) nc.add(cl) elif bottom_point >= x1: embedded_point = [bottom_point, y1] cl = coordinate_line(net, embedded_point, direction) nc.add(cl) else: raise NotImplementedError( "Only works for sampling options 's' and 't' or given float as stepsize" ) if sample_right: extended_bottom_side = modify_direction(bottom_side, 0, [x1, (y2 - y1) + x2]) if isinstance(sampling, Iterable): if len(sampling) != 2: raise ValueError( "You can only specify the sampling of one direction of your domain." " Only works for bounded domains." ) if anchor is not None: raise NotImplementedError( "Options 's' and 't' not compatible with anchor" ) sampled_extended_bottom = netsconv.sample_smooth_domain( extended_bottom_side, sampling, atol=atol ) # Now we need to add the coordinate lines at the points in the # first interval(bottom side), and the intersection points of the # diags with the second interval (right side) for idx in sampled_extended_bottom.domain.traverser: bottom_point = sampled_extended_bottom(idx[0])[0] if bottom_point > x2: # we want to find the points on the "right side" we want to # send coord lines from embedded_point = [x2, bottom_point - x2 + y1] cl = coordinate_line(net, embedded_point, direction) nc.add(cl) elif bottom_point <= x2: # we just need to embed the bottom points embedded_point = [bottom_point, y1] cl = coordinate_line(net, embedded_point, direction) nc.add(cl) elif isinstance(sampling, float) or isinstance(sampling, int): if isinstance(sampling, int): sampling = float(sampling) if anchor is None: bottom_anchor = x2 if anchor is not None: bottom_anchor = anchor[0] + (anchor[1] - y1) sampled_extended_bottom = netsconv.sample_smooth_domain( extended_bottom_side, sampling, anchor=[bottom_anchor], atol=atol ) # CALCULATING sampled points of LEFT side as intersection of the # diaglines in extended bottom side for idx in sampled_extended_bottom.domain.traverser: bottom_point = sampled_extended_bottom(idx[0])[0] if bottom_point > x2: # we want to find the points on the "right side" we want to # send coord lines from embedded_point = [x2, bottom_point - x2 + y1] cl = coordinate_line(net, embedded_point, direction) nc.add(cl) elif bottom_point <= x2: # we just need to embed the bottom points embedded_point = [bottom_point, y1] cl = coordinate_line(net, embedded_point, direction) nc.add(cl) else: raise NotImplementedError( "Only works for sampling options 's' and 't' or given float as stepsize" ) return nc
[docs]def octahedral_grid(net, stepsize, anchor=None, atol=None): """Get diagonal lines in each diagonal direction of the coordinate planes of the domain Parameters ---------- net : ddg.nets.SmoothNet Given 3-dimensional net (must be SmoothNet with rectangular domain). stepsize : int or float Determines the length of the diagonal of the occuring squares. anchor : numpy.ndarray of dimension 3, or None Point inside the domain of the net. The sampling will be chosen such that the point is part of it. atol : float or None (default=None) Is used as tolerance for symmetric sampling. See: sample_interval This function uses the global tolerance defaults if `atol` or `rtol` are set to None. See :py:mod:`ddg.nonexact` for details. Returns ------- list of NetCollections ddg.nets.NetCollection Diagonal lines as a NetCollection. """ nc_xy = NetCollection() nc_yz = NetCollection() nc_xz = NetCollection() dom = ( net.domain.recover() ) # list of intervals with periodictiy information, len(dom)= dimension if len(dom) != 3: raise ValueError("Only works for 3-dim Nets") if anchor is None: anchor = np.array([dom[0][0], dom[1][0], dom[2][0]]) if anchor is not None: # FIRST DIRECTION xy_plane = delete_direction(net.domain, 2) # we sample the z-axis in half the stepsize z_axis = delete_direction(net.domain, 0) z_axis = delete_direction( z_axis, 0 ) # we delete the first and the 2nd direction sampled_z_axis = netsconv.sample_smooth_domain( z_axis, stepsize / 2, anchor=(anchor[2],), atol=atol ) c = 0 for point in sampled_z_axis: def new_fct_2d(a, b, p=point[0], **kwargs): # a point of the sampled z axis is a list of size one, # therefore point[0] return net.fct(a, b, p, **kwargs) new_net = SmoothNet(new_fct_2d, xy_plane) if c % 2 == 0: anch_2d = np.array([anchor[0], anchor[1]]) else: s = stepsize / 2 if anchor[0] - s >= dom[0][0]: anch_2d = np.array([anchor[0] - s, anchor[1]]) else: anch_2d = np.array([anchor[0] + s, anchor[1]]) for line in diagonal_lines( new_net, [1, 1], stepsize, anchor=anch_2d, atol=atol ): nc_xy.add(line) for line in diagonal_lines( new_net, [-1, 1], stepsize, anchor=anch_2d, atol=atol ): nc_xy.add(line) c += 1 # SECOND DIRECTION xz_plane = delete_direction(net.domain, 1) # sampling the y axis y_axis = delete_direction(net.domain, 0) y_axis = delete_direction(y_axis, 1) sampled_y_axis = netsconv.sample_smooth_domain( y_axis, stepsize / 2, anchor=(anchor[1],), atol=atol ) c = 0 for point in sampled_y_axis: def new_fct_2d(a, b, p=point[0], **kwargs): return net.fct(a, p, b, **kwargs) new_net = SmoothNet(new_fct_2d, xz_plane) if c % 2 == 0: anch_2d = np.array([anchor[0], anchor[2]]) else: s = stepsize / 2 if anchor[0] - s >= dom[0][0]: anch_2d = np.array([anchor[0] - s, anchor[2]]) else: anch_2d = np.array([anchor[0] + s, anchor[2]]) for line in diagonal_lines( new_net, [1, 1], stepsize, anchor=anch_2d, atol=atol ): nc_xz.add(line) for line in diagonal_lines( new_net, [-1, 1], stepsize, anchor=anch_2d, atol=atol ): nc_xz.add(line) c += 1 # THIRD DIRECTION yz_plane = delete_direction(net.domain, 0) # sampling the x axis x_axis = delete_direction(net.domain, 1) x_axis = delete_direction(x_axis, 1) # deleting second and third dir sampled_x_axis = netsconv.sample_smooth_domain( x_axis, stepsize / 2, anchor=(anchor[0],), atol=atol ) c = 0 for point in sampled_x_axis: def new_fct_2d(a, b, p=point[0], **kwargs): return net.fct(p, a, b, **kwargs) new_net = SmoothNet(new_fct_2d, yz_plane) if c % 2 == 0: anch_2d = np.array([anchor[1], anchor[2]]) else: s = stepsize / 2 if anchor[0] - s >= dom[0][0]: anch_2d = np.array([anchor[1] - s, anchor[2]]) else: anch_2d = np.array([anchor[1] + s, anchor[2]]) for line in diagonal_lines( new_net, [1, 1], stepsize, anchor=anch_2d, atol=atol ): nc_yz.add(line) for line in diagonal_lines( new_net, [-1, 1], stepsize, anchor=anch_2d, atol=atol ): nc_yz.add(line) c += 1 return [nc_yz, nc_xz, nc_xy]
[docs]@domain_util def shrink_domain(domain, amount): """ Shrinks domain by a given amount from the left and right. Parameters ---------- domain : ddg.nets.DiscreteRectangularDomain or SmoothRectangularDomain Domain to shrink, or a net to take the domain from. Discrete domains can only be shrunk by integer values. amount : list or float | Specifies how much is taken away in each direction from the domain: | * float/int - value to take away from left and right in every direction | * list - every entry in the list corresponds to one direction, | and may be None, int, or a tuple of int or None. Returns ------- ddg.nets.Domain Shrunken domain, if the domain was given. ddg.nets.Net The input discrete net containing the new shrunken domain, if the net was given. Notes ----- * Only bounded directions will be shrunken. * Periodicity is inherited only for the directions that are not shrunken. Raises ------ TypeError if domain was discrete, but non-integer amount was passed """ # noqa: E501 if not isinstance(amount, Iterable): amount = [amount for _ in range(domain.dimension)] if isinstance(domain, DiscreteDomain) and any( not isinstance(i, int) for i in amount ): raise TypeError("Discrete domains can only be shrunk by integer values.") shrunken_intervals = [] for i, (s, interval) in enumerate(zip(amount, domain.intervals)): if s is None or not _is_bounded_interval(interval): shrunken_intervals.append(interval) else: if len(np.shape(s)) == 0: s = [s, s] a, b = interval p = i in domain.periodicity s0, s1 = s if s0 is not None: a = a + s0 if s1 is not None: b = b - s1 shrunken_intervals.append([a, b, p]) return shrunken_intervals
[docs]@domain_util def bound_domain(domain, bounding): """ Parameters ---------- domain : ddg.nets.Smooth/DiscreteRectangularDomain or ddg.nets.Smooth/DiscreteNet Given (unbounded) rectangular domain to bound, or net whose domain is to be bound bounding : int/float | Bounding range. | If an interval (a,b) in the domain is unbounded, | we bound it in the smooth case by | (1) [a, a+bounding], if a > -inf | (2) [b-bounding, b], if b < inf, or | (3) [-bounding/2, bounding/2] if a == -inf and b == inf. | | In the discrete case, bounding is the total amount of points chosen in | an unbounded direction. | (1) [a, a+bounding-1], if a > -inf | (2) [b-bounding+1, b], if b < inf, or | (3) [1-bounding/2, bounding/2] if a == -inf and b == inf. Returns ------- ddg.nets.Smooth/DiscreteRectangularDomain or ddg.nets.Smooth/DiscreteNet Will be the same type as domain, e.g. a discrete/smooth rectangular domain, if domain was discrete/smooth rectangular domain. """ # noqa: E501 # Empty domains get handled by the decorator if domain.bounded: # domain.recover() doesn't work because of SmoothInterval and # DiscreteInterval return [d + [i in domain.periodicity] for i, d in enumerate(domain.intervals)] discrete = 0 if isinstance(domain, DiscreteDomain): discrete = 1 intervals = [] for idx, el in enumerate(domain): if _is_bounded_interval(el): intervals.append(el + [idx in domain.periodicity]) else: a, b = el if a > -np.inf: intervals.append([a, a + bounding - discrete]) elif b < np.inf: intervals.append([b - bounding + discrete, b]) else: n, re = divmod(bounding, 2) intervals.append([discrete - re - n, n]) return intervals
[docs]@domain_util def create_subdomain(domain, subdomain): """ Create subdomain of a given discrete domain. WARNING: To assure that the output is always a subdomain of the given domain, subdomain will be intersected with the domain. Parameters ---------- domain : ddg.nets.DiscreteRectangularDomain Given domain to reduce to a subdomain, or net. subdomain : list, or DiscreteDomain | Each entry in the list corresponds to a direction: | * int - number of values to take | * tuple - subdomain in given direction | If a DiscreteDomain is given, it will just be returned as is. Returns ------- ddg.datastrucuters.nets.domain.DiscreteRectangularDomain Subdomain. ddg.nets.Net A discrete net containing the new shrunken domain, if the net was given. """ intervals = [] if subdomain is None: subdomain = [None] * domain.dimension for idx, (el, interval) in enumerate(zip(subdomain, domain)): if el is None or el == interval: intervals.append(interval + [idx in domain.periodicity]) else: ela, elb = el ia, ib = interval intervals.append([max(ela, ia), min(elb, ib)]) return intervals
[docs]@net_util def continue_by_reflection(net, direction, reflection, invertdirection=False): """ Continue a net by reflection. Given a net, direction and reflection, the function returns a net with an expanded domain in direction direction. Points outside the domain of the orignal domain will be mapped to a point corresponding to it inside of the domain before the new net evaluates first the function of the original net before applying the reflection. Examples -------- >>> import numpy as np >>> from ddg.nets import SmoothNet >>> from ddg.nets import continue_by_reflection >>> net = SmoothNet(lambda *a: np.array(a), [[0, 2]] * 2) >>> reflectionnet = continue_by_reflection(net, 0, 0) >>> reflectionnet(3, 0) array([-1, 0]) Parameters ---------- net : ddg.nets.SmoothNet or DiscreteNet Net to continue by reflection direction : int Direction in which the domain will be expanded. Points outside the original domain will be mapped to the point corresponding to it and the reflection will be applied after the function was evalued reflection : int, Iterable, function | If given an integer or an Iterable containing integers, reflection | determines the coordinates changing sign after the original net is | evalued at the point corresponding to the one given outside the | original domain, e.g. for a point p outside of the original domain | corresponding to p~ and reflect the function given by the int/Iterable | | newNet(p) == reflect(net(p~)) | | If a function is given, it will be used instead. | | WARNING: | In case a function is given, it will not be checked if it can | accept the output of the net or not. invertdirection : bool, optional If set to False, the domain will be expanded in positive direction If set to True, the domain will be expanded in negative direction Returns ------- ddg.nets.Net Same type as net Raises ------ IndexError If direction is not in range of domain dimension TypeError If reflection is neither an int, Iterable or function """ domain = net.domain if direction >= domain.dimension: raise IndexError("Dimension of net domain is < {}.".format(direction + 1)) if not callable(reflection): if isinstance(reflection, Iterable) and all( [isinstance(x, int) for x in reflection] ): def rflct(*args, idx=reflection): nargs = np.array(args) reflect = np.array([-1 if i in idx else 1 for i in range(len(nargs))]) return nargs * reflect elif isinstance(reflection, int): def rflct(*args, idx=reflection): nargs = np.array(args) reflect = np.array([-1 if i == idx else 1 for i in range(len(nargs))]) return nargs * reflect else: msg = "Does not support {} as reflection.".format(type(reflection)) raise TypeError(msg) reflection = rflct if not invertdirection: intervals = [] for idx, el in enumerate(domain): if idx != direction: intervals.append(el + [idx in domain.periodicity]) else: a, b = el intervals.append([a, 2 * b - a]) def fct(*args, **kwargs): bound = domain.intervals[direction][1] if args in domain: return net(*args, **kwargs) else: nargs = np.array(args) nargs[direction] = 2 * bound - nargs[direction] return reflection(*net(*nargs)) else: intervals = [] for idx, el in enumerate(domain): if idx != direction: intervals.append(el + [idx in domain.periodicity]) else: a, b = el intervals.append([2 * a - b, b]) def fct(*args, **kwargs): bound = domain.intervals[direction][0] if args in domain: return net(*args, **kwargs) else: nargs = np.array(args) nargs[direction] = 2 * bound - nargs[direction] return reflection(*net(*nargs)) return (fct, intervals)
[docs]@net_util def concatenate(net, fct): """ Create a net which's function is the conatenation of the function of the original net and fct and has the same domain as the original. Parameters ---------- net : ddg.nets.Net or subclass of it Net with compatible function for fct fct : function Function to concatenate with the net function Returns ------- ddg.nets.Net Exact type is the same as given net """ def nfct(*args, **kwargs): return fct(net(*args)) intervals = net.domain.intervals return (nfct, intervals)
[docs]def evaluate(net): """ Evaluate a discrete net on its entire domain. This should NOT replace net[net.domain]. Instead use this function if the function of a net has effects outside it, i.e. when a net creates blender object and link them to the scene on its own. Parameters ---------- net : ddg.nets.DiscreteNet or subclass of it Returns ------- 0 If successfull Raises ------ TypeError if net was not a (subclass of) ddg.nets.DiscreteNet """ if not isinstance(net, DiscreteNet): raise TypeError("Only works for discrete nets.") return net.evaluate()
[docs]@domain_util def delete_direction(domain, direction): """ Delete one direction of a domain. Parameters ---------- domain: SmoothRectangularDomain or Discrete RectangularDomain may be parametrized direction: int Returns ------- domain of same class of one dimension less """ intdict = {} idx = 0 for interval in domain: # creating dictionary -> index:interval intdict[idx] = interval idx += 1 del intdict[direction] less_intervals = list(intdict.values()) # conv dictionary of intervals to list return less_intervals
[docs]@domain_util def modify_direction(domain, direction, new_interval): intdict = {} idx = 0 for interval in domain: # creating dictionary -> index:interval intdict[idx] = interval idx += 1 intdict[direction] = new_interval new_intervals = list(intdict.values()) # conv dictionary of intervals to list return new_intervals
[docs]@applicable_to_netcollection def cut_bounding_box(curve, distances=(1.0, 1.0, 1.0), center=(0.0, 0.0, 0.0)): """Cuts the curve within the given bounding box. * only works for curves with infinite domain * works best for curves with small sample size (no additional points on the boundary of the bounding_box are set) The box is given by:: {x in R^3 | - distances[i] < <x - center, e_i> < distances[i]}. A picture of the box:: y-axis │ ┌───────────▲───────────┐ │ │ d_0│ │ d_0 = distances[0] │ │ │ d_1 │ d_1 = distances[1] │ │ └───────────► │ │ center │ │ │ │ │ └───────────────────────┘ └────────────────────────────► z-axis points at eye x-axis Parameters ---------- curve : ddg.nets.DiscreteCurve curve to be cut distances : sequence of three floats (default=(1.0,1.0,1.0)) Distances of planes parallel to the coordinate planes to the given center. center: sequence of three floats (default=(0.0,0.0,0.0)) Center of the bounding box. Raises ------ ValueError If the curves domain is infinite. If `len(distances) != 3`. If `len(center) != 3`. Returns ------- ddg.nets.NetCollection NetCollection of the remaining segments of the curve """ if not _is_bounded_interval(curve.domain.interval): raise ValueError( "The curves domain has to be bounded, the given curves domain is " "unbounded:", curve.domain.interval, ) if len(distances) != 3: raise ValueError(f"{distances = } must have length 3.") if len(center) != 3: raise ValueError(f"{center = } must have length 3.") center = np.array(center) bounding_box_abs = np.array([np.abs(x) for x in distances]) interval_collection = [] out = True for i in range(curve.domain.interval[0], curve.domain.interval[1] + 1): vec = np.array(curve.fct(i)) if ( not (np.array(vec) < center + bounding_box_abs).all() or not (np.array(vec) > center - bounding_box_abs).all() ): out = True else: if out: interval_collection.append([i, i]) out = False else: interval_collection[-1][-1] = i curve_segments = [] if curve.domain.periodicity: if len(interval_collection) >= 2: I1 = interval_collection.pop(0) I2 = interval_collection.pop() if I1[0] == curve.domain.interval[0] and I2[1] == curve.domain.interval[1]: I2[1] += I1[1] + 1 def periodicity_fct(x, b): if x <= b: return curve.fct(x) else: return curve.fct(x - b - 1) curve_segments.append(ddg.nets.DiscreteCurve(periodicity_fct, I2)) for interval in interval_collection: curve_segments.append(ddg.nets.DiscreteCurve(curve.fct, interval)) if len(curve_segments) == 1: net = curve_segments[0] else: net = NetCollection(curve_segments) return net