Source code for yt.utilities.math_utils

import math

import numpy as np

from yt.units.yt_array import YTArray

prec_accum = {
    int: np.int64,
    np.int8: np.int64,
    np.int16: np.int64,
    np.int32: np.int64,
    np.int64: np.int64,
    np.uint8: np.uint64,
    np.uint16: np.uint64,
    np.uint32: np.uint64,
    np.uint64: np.uint64,
    float: np.float64,
    np.float16: np.float64,
    np.float32: np.float64,
    np.float64: np.float64,
    complex: np.complex128,
    np.complex64: np.complex128,
    np.complex128: np.complex128,
    np.dtype("int"): np.int64,
    np.dtype("int8"): np.int64,
    np.dtype("int16"): np.int64,
    np.dtype("int32"): np.int64,
    np.dtype("int64"): np.int64,
    np.dtype("uint8"): np.uint64,
    np.dtype("uint16"): np.uint64,
    np.dtype("uint32"): np.uint64,
    np.dtype("uint64"): np.uint64,
    np.dtype("float"): np.float64,
    np.dtype("float16"): np.float64,
    np.dtype("float32"): np.float64,
    np.dtype("float64"): np.float64,
    np.dtype("complex"): np.complex128,
    np.dtype("complex64"): np.complex128,
    np.dtype("complex128"): np.complex128,
}


[docs] def periodic_position(pos, ds): r"""Assuming periodicity, find the periodic position within the domain. Parameters ---------- pos : array An array of floats. ds : ~yt.data_objects.static_output.Dataset A simulation static output. Examples -------- >>> a = np.array([1.1, 0.5, 0.5]) >>> data = {"Density": np.ones([32, 32, 32])} >>> ds = load_uniform_grid(data, [32, 32, 32], 1.0) >>> ppos = periodic_position(a, ds) >>> ppos array([ 0.1, 0.5, 0.5]) """ off = (pos - ds.domain_left_edge) % ds.domain_width return ds.domain_left_edge + off
[docs] def periodic_dist(a, b, period, periodicity=(True, True, True)): r"""Find the Euclidean periodic distance between two sets of points. Parameters ---------- a : array or list Either an ndim long list of coordinates corresponding to a single point or an (ndim, npoints) list of coordinates for many points in space. b : array of list Either an ndim long list of coordinates corresponding to a single point or an (ndim, npoints) list of coordinates for many points in space. period : float or array or list If the volume is symmetrically periodic, this can be a single float, otherwise an array or list of floats giving the periodic size of the volume for each dimension. periodicity : An ndim-element tuple of booleans If an entry is true, the domain is assumed to be periodic along that direction. Examples -------- >>> a = [0.1, 0.1, 0.1] >>> b = [0.9, 0, 9, 0.9] >>> period = 1.0 >>> dist = periodic_dist(a, b, 1.0) >>> dist 0.346410161514 """ a = np.array(a) b = np.array(b) period = np.array(period) if period.size == 1: period = np.array([period, period, period]) if a.shape != b.shape: raise RuntimeError("Arrays must be the same shape.") if period.shape != a.shape and len(a.shape) > 1: n_tup = tuple(1 for i in range(a.ndim - 1)) period = np.tile(np.reshape(period, (a.shape[0],) + n_tup), (1,) + a.shape[1:]) elif len(a.shape) == 1: a = np.reshape(a, (a.shape[0],) + (1, 1)) b = np.reshape(b, (a.shape[0],) + (1, 1)) period = np.reshape(period, (a.shape[0],) + (1, 1)) c = np.empty((2,) + a.shape, dtype="float64") c[0, :] = np.abs(a - b) p_directions = [i for i, p in enumerate(periodicity) if p] np_directions = [i for i, p in enumerate(periodicity) if not p] for d in p_directions: c[1, d, :] = period[d, :] - np.abs(a - b)[d, :] for d in np_directions: c[1, d, :] = c[0, d, :] d = np.amin(c, axis=0) ** 2 r2 = d.sum(axis=0) if r2.size == 1: return np.sqrt(r2[0, 0]) return np.sqrt(r2)
[docs] def periodic_ray(start, end, left=None, right=None): """ periodic_ray(start, end, left=None, right=None) Break up periodic ray into non-periodic segments. Accepts start and end points of periodic ray as YTArrays. Accepts optional left and right edges of periodic volume as YTArrays. Returns a list of lists of coordinates, where each element of the top-most list is a 2-list of start coords and end coords of the non-periodic ray: [[[x0start,y0start,z0start], [x0end, y0end, z0end]], [[x1start,y1start,z1start], [x1end, y1end, z1end]], ...,] Parameters ---------- start : array The starting coordinate of the ray. end : array The ending coordinate of the ray. left : optional, array The left corner of the periodic domain. If not given, an array of zeros with same size as the starting coordinate us used. right : optional, array The right corner of the periodic domain. If not given, an array of ones with same size as the starting coordinate us used. Examples -------- >>> import yt >>> start = yt.YTArray([0.5, 0.5, 0.5]) >>> end = yt.YTArray([1.25, 1.25, 1.25]) >>> periodic_ray(start, end) [ [ YTArray([0.5, 0.5, 0.5]) (dimensionless), YTArray([1., 1., 1.]) (dimensionless) ], [ YTArray([0., 0., 0.]) (dimensionless), YTArray([0.25, 0.25, 0.25]) (dimensionless) ] ] """ if left is None: left = np.zeros(start.shape) if right is None: right = np.ones(start.shape) dim = right - left vector = end - start wall = np.zeros_like(start) close = np.zeros(start.shape, dtype=object) left_bound = vector < 0 right_bound = vector > 0 no_bound = vector == 0.0 bound = vector != 0.0 wall[left_bound] = left[left_bound] close[left_bound] = np.max wall[right_bound] = right[right_bound] close[right_bound] = np.min wall[no_bound] = np.inf close[no_bound] = np.min segments = [] this_start = start.copy() this_end = end.copy() t = 0.0 tolerance = 1e-6 while t < 1.0 - tolerance: hit_left = (this_start <= left) & (vector < 0) if (hit_left).any(): this_start[hit_left] += dim[hit_left] this_end[hit_left] += dim[hit_left] hit_right = (this_start >= right) & (vector > 0) if (hit_right).any(): this_start[hit_right] -= dim[hit_right] this_end[hit_right] -= dim[hit_right] nearest = vector.unit_array * np.array( [close[q]([this_end[q], wall[q]]) for q in range(start.size)] ) dt = ((nearest - this_start) / vector)[bound].min() now = this_start + vector * dt close_enough = np.abs(now - nearest) / np.abs(vector.max()) < 1e-10 now[close_enough] = nearest[close_enough] segments.append([this_start.copy(), now.copy()]) this_start = now.copy() t += dt return segments
[docs] def euclidean_dist(a, b): r"""Find the Euclidean distance between two points. Parameters ---------- a : array or list Either an ndim long list of coordinates corresponding to a single point or an (ndim, npoints) list of coordinates for many points in space. b : array or list Either an ndim long list of coordinates corresponding to a single point or an (ndim, npoints) list of coordinates for many points in space. Examples -------- >>> a = [0.1, 0.1, 0.1] >>> b = [0.9, 0, 9, 0.9] >>> period = 1.0 >>> dist = euclidean_dist(a, b) >>> dist 1.38564064606 """ a = np.array(a) b = np.array(b) if a.shape != b.shape: RuntimeError("Arrays must be the same shape.") c = a.copy() np.subtract(c, b, c) np.power(c, 2, c) c = c.sum(axis=0) if isinstance(c, np.ndarray): np.sqrt(c, c) else: # This happens if a and b only have one entry. c = math.sqrt(c) return c
[docs] def rotate_vector_3D(a, dim, angle): r"""Rotates the elements of an array around an axis by some angle. Given an array of 3D vectors a, this rotates them around a coordinate axis by a clockwise angle. An alternative way to think about it is the coordinate axes are rotated counterclockwise, which changes the directions of the vectors accordingly. Parameters ---------- a : array An array of 3D vectors with dimension Nx3. dim : integer A integer giving the axis around which the vectors will be rotated. (x, y, z) = (0, 1, 2). angle : float The angle in radians through which the vectors will be rotated clockwise. Examples -------- >>> a = np.array([[1, 1, 0], [1, 0, 1], [0, 1, 1], [1, 1, 1], [3, 4, 5]]) >>> b = rotate_vector_3D(a, 2, np.pi / 2) >>> print(b) [[ 1.00000000e+00 -1.00000000e+00 0.00000000e+00] [ 6.12323400e-17 -1.00000000e+00 1.00000000e+00] [ 1.00000000e+00 6.12323400e-17 1.00000000e+00] [ 1.00000000e+00 -1.00000000e+00 1.00000000e+00] [ 4.00000000e+00 -3.00000000e+00 5.00000000e+00]] """ mod = False if len(a.shape) == 1: mod = True a = np.array([a]) if a.shape[1] != 3: raise ValueError("The second dimension of the array a must be == 3!") if dim == 0: R = np.array( [ [1, 0, 0], [0, np.cos(angle), np.sin(angle)], [0, -np.sin(angle), np.cos(angle)], ] ) elif dim == 1: R = np.array( [ [np.cos(angle), 0, -np.sin(angle)], [0, 1, 0], [np.sin(angle), 0, np.cos(angle)], ] ) elif dim == 2: R = np.array( [ [np.cos(angle), np.sin(angle), 0], [-np.sin(angle), np.cos(angle), 0], [0, 0, 1], ] ) else: raise ValueError("dim must be 0, 1, or 2!") if mod: return np.dot(R, a.T).T[0] else: return np.dot(R, a.T).T
[docs] def modify_reference_frame(CoM, L, P=None, V=None): r"""Rotates and translates data into a new reference frame to make calculations easier. This is primarily useful for calculations of halo data. The data is translated into the center of mass frame. Next, it is rotated such that the angular momentum vector for the data is aligned with the z-axis. Put another way, if one calculates the angular momentum vector on the data that comes out of this function, it will always be along the positive z-axis. If the center of mass is re-calculated, it will be at the origin. Parameters ---------- CoM : array The center of mass in 3D. L : array The angular momentum vector. Optional -------- P : array The positions of the data to be modified (i.e. particle or grid cell positions). The array should be Nx3. V : array The velocities of the data to be modified (i.e. particle or grid cell velocities). The array should be Nx3. Returns ------- L : array The angular momentum vector equal to [0, 0, 1] modulo machine error. P : array The modified positional data. Only returned if P is not None V : array The modified velocity data. Only returned if V is not None Examples -------- >>> CoM = np.array([0.5, 0.5, 0.5]) >>> L = np.array([1, 0, 0]) >>> P = np.array([[1, 0.5, 0.5], [0, 0.5, 0.5], [0.5, 0.5, 0.5], [0, 0, 0]]) >>> V = p.copy() >>> LL, PP, VV = modify_reference_frame(CoM, L, P, V) >>> LL array([ 6.12323400e-17, 0.00000000e+00, 1.00000000e+00]) >>> PP array([[ 3.06161700e-17, 0.00000000e+00, 5.00000000e-01], [ -3.06161700e-17, 0.00000000e+00, -5.00000000e-01], [ 0.00000000e+00, 0.00000000e+00, 0.00000000e+00], [ 5.00000000e-01, -5.00000000e-01, -5.00000000e-01]]) >>> VV array([[ -5.00000000e-01, 5.00000000e-01, 1.00000000e+00], [ -5.00000000e-01, 5.00000000e-01, 3.06161700e-17], [ -5.00000000e-01, 5.00000000e-01, 5.00000000e-01], [ 0.00000000e+00, 0.00000000e+00, 0.00000000e+00]]) """ # First translate the positions to center of mass reference frame. if P is not None: P = P - CoM # is the L vector pointing in the Z direction? if np.inner(L[:2], L[:2]) == 0.0: # the reason why we need to explicitly check for the above # is that formula is used in denominator # this just checks if we need to flip the z axis or not if L[2] < 0.0: # this is just a simple flip in direction of the z axis if P is not None: P = -P if V is not None: V = -V # return the values if V is None and P is not None: return L, P elif P is None and V is not None: return L, V else: return L, P, V # Normal vector is not aligned with simulation Z axis # Therefore we are going to have to apply a rotation # Now find the angle between modified L and the x-axis. LL = L.copy() LL[2] = 0.0 theta = np.arccos(np.inner(LL, [1.0, 0.0, 0.0]) / np.inner(LL, LL) ** 0.5) if L[1] < 0.0: theta = -theta # Now rotate all the position, velocity, and L vectors by this much around # the z axis. if P is not None: P = rotate_vector_3D(P, 2, theta) if V is not None: V = rotate_vector_3D(V, 2, theta) L = rotate_vector_3D(L, 2, theta) # Now find the angle between L and the z-axis. theta = np.arccos(np.inner(L, [0.0, 0.0, 1.0]) / np.inner(L, L) ** 0.5) # This time we rotate around the y axis. if P is not None: P = rotate_vector_3D(P, 1, theta) if V is not None: V = rotate_vector_3D(V, 1, theta) L = rotate_vector_3D(L, 1, theta) # return the values if V is None and P is not None: return L, P elif P is None and V is not None: return L, V else: return L, P, V
[docs] def compute_rotational_velocity(CoM, L, P, V): r"""Computes the rotational velocity for some data around an axis. This is primarily for halo computations. Given some data, this computes the circular rotational velocity of each point (particle) in reference to the axis defined by the angular momentum vector. This is accomplished by converting the reference frame of the center of mass of the halo. Parameters ---------- CoM : array The center of mass in 3D. L : array The angular momentum vector. P : array The positions of the data to be modified (i.e. particle or grid cell positions). The array should be Nx3. V : array The velocities of the data to be modified (i.e. particle or grid cell velocities). The array should be Nx3. Returns ------- v : array An array N elements long that gives the circular rotational velocity for each datum (particle). Examples -------- >>> CoM = np.array([0, 0, 0]) >>> L = np.array([0, 0, 1]) >>> P = np.array([[1, 0, 0], [1, 1, 1], [0, 0, 1], [1, 1, 0]]) >>> V = np.array([[0, 1, 10], [-1, -1, -1], [1, 1, 1], [1, -1, -1]]) >>> circV = compute_rotational_velocity(CoM, L, P, V) >>> circV array([ 1. , 0. , 0. , 1.41421356]) """ # First we translate into the simple coordinates. L, P, V = modify_reference_frame(CoM, L, P, V) # Find the vector in the plane of the galaxy for each position point # that is perpendicular to the radial vector. radperp = np.cross([0, 0, 1], P) # Find the component of the velocity along the radperp vector. # Unf., I don't think there's a better way to do this. res = np.empty(V.shape[0], dtype="float64") for i, rp in enumerate(radperp): temp = np.dot(rp, V[i]) / np.dot(rp, rp) * rp res[i] = np.dot(temp, temp) ** 0.5 return res
[docs] def compute_parallel_velocity(CoM, L, P, V): r"""Computes the parallel velocity for some data around an axis. This is primarily for halo computations. Given some data, this computes the velocity component along the angular momentum vector. This is accomplished by converting the reference frame of the center of mass of the halo. Parameters ---------- CoM : array The center of mass in 3D. L : array The angular momentum vector. P : array The positions of the data to be modified (i.e. particle or grid cell positions). The array should be Nx3. V : array The velocities of the data to be modified (i.e. particle or grid cell velocities). The array should be Nx3. Returns ------- v : array An array N elements long that gives the parallel velocity for each datum (particle). Examples -------- >>> CoM = np.array([0, 0, 0]) >>> L = np.array([0, 0, 1]) >>> P = np.array([[1, 0, 0], [1, 1, 1], [0, 0, 1], [1, 1, 0]]) >>> V = np.array([[0, 1, 10], [-1, -1, -1], [1, 1, 1], [1, -1, -1]]) >>> paraV = compute_parallel_velocity(CoM, L, P, V) >>> paraV array([10, -1, 1, -1]) """ # First we translate into the simple coordinates. L, P, V = modify_reference_frame(CoM, L, P, V) # And return just the z-axis velocities. return V[:, 2]
[docs] def compute_radial_velocity(CoM, L, P, V): r"""Computes the radial velocity for some data around an axis. This is primarily for halo computations. Given some data, this computes the radial velocity component for the data. This is accomplished by converting the reference frame of the center of mass of the halo. Parameters ---------- CoM : array The center of mass in 3D. L : array The angular momentum vector. P : array The positions of the data to be modified (i.e. particle or grid cell positions). The array should be Nx3. V : array The velocities of the data to be modified (i.e. particle or grid cell velocities). The array should be Nx3. Returns ------- v : array An array N elements long that gives the radial velocity for each datum (particle). Examples -------- >>> CoM = np.array([0, 0, 0]) >>> L = np.array([0, 0, 1]) >>> P = np.array([[1, 0, 0], [1, 1, 1], [0, 0, 1], [1, 1, 0]]) >>> V = np.array([[0, 1, 10], [-1, -1, -1], [1, 1, 1], [1, -1, -1]]) >>> radV = compute_radial_velocity(CoM, L, P, V) >>> radV array([ 1. , 1.41421356 , 0. , 0.]) """ # First we translate into the simple coordinates. L, P, V = modify_reference_frame(CoM, L, P, V) # We find the tangential velocity by dotting the velocity vector # with the cylindrical radial vector for this point. # Unf., I don't think there's a better way to do this. P[:, 2] = 0 res = np.empty(V.shape[0], dtype="float64") for i, rad in enumerate(P): temp = np.dot(rad, V[i]) / np.dot(rad, rad) * rad res[i] = np.dot(temp, temp) ** 0.5 return res
[docs] def compute_cylindrical_radius(CoM, L, P, V): r"""Compute the radius for some data around an axis in cylindrical coordinates. This is primarily for halo computations. Given some data, this computes the cylindrical radius for each point. This is accomplished by converting the reference frame of the center of mass of the halo. Parameters ---------- CoM : array The center of mass in 3D. L : array The angular momentum vector. P : array The positions of the data to be modified (i.e. particle or grid cell positions). The array should be Nx3. V : array The velocities of the data to be modified (i.e. particle or grid cell velocities). The array should be Nx3. Returns ------- cyl_r : array An array N elements long that gives the radial velocity for each datum (particle). Examples -------- >>> CoM = np.array([0, 0, 0]) >>> L = np.array([0, 0, 1]) >>> P = np.array([[1, 0, 0], [1, 1, 1], [0, 0, 1], [1, 1, 0]]) >>> V = np.array([[0, 1, 10], [-1, -1, -1], [1, 1, 1], [1, -1, -1]]) >>> cyl_r = compute_cylindrical_radius(CoM, L, P, V) >>> cyl_r array([ 1. , 1.41421356, 0. , 1.41421356]) """ # First we translate into the simple coordinates. L, P, V = modify_reference_frame(CoM, L, P, V) # Demote all the positions to the z=0 plane, which makes the distance # calculation very easy. P[:, 2] = 0 return np.sqrt((P * P).sum(axis=1))
[docs] def ortho_find(vec1): r"""Find two complementary orthonormal vectors to a given vector. For any given non-zero vector, there are infinite pairs of vectors orthonormal to it. This function gives you one arbitrary pair from that set along with the normalized version of the original vector. Parameters ---------- vec1 : array_like An array or list to represent a 3-vector. Returns ------- vec1 : array The original 3-vector array after having been normalized. vec2 : array A 3-vector array which is orthonormal to vec1. vec3 : array A 3-vector array which is orthonormal to vec1 and vec2. Raises ------ ValueError If input vector is the zero vector. Notes ----- Our initial vector is `vec1` which consists of 3 components: `x1`, `y1`, and `z1`. ortho_find determines a vector, `vec2`, which is orthonormal to `vec1` by finding a vector which has a zero-value dot-product with `vec1`. .. math:: vec1 \cdot vec2 = x_1 x_2 + y_1 y_2 + z_1 z_2 = 0 As a starting point, we arbitrarily choose `vec2` to have `x2` = 1, `y2` = 0: .. math:: vec1 \cdot vec2 = x_1 + (z_1 z_2) = 0 \rightarrow z_2 = -(x_1 / z_1) Of course, this will fail if `z1` = 0, in which case, let's say use `z2` = 1 and `x2` = 0: .. math:: \rightarrow y_2 = -(z_1 / y_1) Similarly, if `y1` = 0, this case will fail, in which case we use `y2` = 1 and `z2` = 0: .. math:: \rightarrow x_2 = -(y_1 / x_1) Since we don't allow `vec1` to be zero, all cases are accounted for. Producing `vec3`, the complementary orthonormal vector to `vec1` and `vec2` is accomplished by simply taking the cross product of `vec1` and `vec2`. Examples -------- >>> a = [1.0, 2.0, 3.0] >>> a, b, c = ortho_find(a) >>> a array([ 0.26726124, 0.53452248, 0.80178373]) >>> b array([ 0.9486833 , 0. , -0.31622777]) >>> c array([-0.16903085, 0.84515425, -0.50709255]) """ vec1 = np.array(vec1, dtype=np.float64) # Normalize norm = np.sqrt(np.vdot(vec1, vec1)) if norm == 0: raise ValueError("Zero vector used as input.") vec1 /= norm x1 = vec1[0] y1 = vec1[1] z1 = vec1[2] if z1 != 0: x2 = 1.0 y2 = 0.0 z2 = -(x1 / z1) norm2 = (1.0 + z2**2.0) ** (0.5) elif y1 != 0: x2 = 0.0 z2 = 1.0 y2 = -(z1 / y1) norm2 = (1.0 + y2**2.0) ** (0.5) else: y2 = 1.0 z2 = 0.0 x2 = -(y1 / x1) norm2 = (1.0 + z2**2.0) ** (0.5) vec2 = np.array([x2, y2, z2]) vec2 /= norm2 vec3 = np.cross(vec1, vec2) return vec1, vec2, vec3
[docs] def quartiles(a, axis=None, out=None, overwrite_input=False): """ Compute the quartile values (25% and 75%) along the specified axis in the same way that the numpy.median calculates the median (50%) value alone a specified axis. Check numpy.median for details, as it is virtually the same algorithm. Returns an array of the quartiles of the array elements [lower quartile, upper quartile]. Parameters ---------- a : array_like Input array or object that can be converted to an array. axis : {None, int}, optional Axis along which the quartiles are computed. The default (axis=None) is to compute the quartiles along a flattened version of the array. out : ndarray, optional Alternative output array in which to place the result. It must have the same shape and buffer length as the expected output, but the type (of the output) will be cast if necessary. overwrite_input : {False, True}, optional If True, then allow use of memory of input array (a) for calculations. The input array will be modified by the call to quartiles. This will save memory when you do not need to preserve the contents of the input array. Treat the input as undefined, but it will probably be fully or partially sorted. Default is False. Note that, if `overwrite_input` is True and the input is not already an ndarray, an error will be raised. Returns ------- quartiles : ndarray A new 2D array holding the result (unless `out` is specified, in which case that array is returned instead). If the input contains integers, or floats of smaller precision than 64, then the output data-type is float64. Otherwise, the output data-type is the same as that of the input. See Also -------- numpy.median, numpy.mean, numpy.percentile Notes ----- Given a vector V of length N, the quartiles of V are the 25% and 75% values of a sorted copy of V, ``V_sorted`` - i.e., ``V_sorted[(N-1)/4]`` and ``3*V_sorted[(N-1)/4]``, when N is odd. When N is even, it is the average of the two values bounding these values of ``V_sorted``. Examples -------- >>> a = np.arange(100).reshape(10, 10) >>> a array([[ 0, 1, 2, 3, 4, 5, 6, 7, 8, 9], [10, 11, 12, 13, 14, 15, 16, 17, 18, 19], [20, 21, 22, 23, 24, 25, 26, 27, 28, 29], [30, 31, 32, 33, 34, 35, 36, 37, 38, 39], [40, 41, 42, 43, 44, 45, 46, 47, 48, 49], [50, 51, 52, 53, 54, 55, 56, 57, 58, 59], [60, 61, 62, 63, 64, 65, 66, 67, 68, 69], [70, 71, 72, 73, 74, 75, 76, 77, 78, 79], [80, 81, 82, 83, 84, 85, 86, 87, 88, 89], [90, 91, 92, 93, 94, 95, 96, 97, 98, 99]]) >>> mu.quartiles(a) array([ 24.5, 74.5]) >>> mu.quartiles(a, axis=0) array([[ 15., 16., 17., 18., 19., 20., 21., 22., 23., 24.], [ 65., 66., 67., 68., 69., 70., 71., 72., 73., 74.]]) >>> mu.quartiles(a, axis=1) array([[ 1.5, 11.5, 21.5, 31.5, 41.5, 51.5, 61.5, 71.5, 81.5, 91.5], [ 6.5, 16.5, 26.5, 36.5, 46.5, 56.5, 66.5, 76.5, 86.5, 96.5]]) """ if overwrite_input: if axis is None: a_sorted = sorted(a.ravel()) else: a.sort(axis=axis) a_sorted = a else: a_sorted = np.sort(a, axis=axis) if axis is None: axis = 0 indexer = [slice(None)] * a_sorted.ndim indices = [int(a_sorted.shape[axis] / 4), int(a_sorted.shape[axis] * 0.75)] result = [] for index in indices: if a_sorted.shape[axis] % 2 == 1: # index with slice to allow mean (below) to work indexer[axis] = slice(index, index + 1) else: indexer[axis] = slice(index - 1, index + 1) # special cases for small arrays if a_sorted.shape[axis] == 2: # index with slice to allow mean (below) to work indexer[axis] = slice(index, index + 1) # Use mean in odd and even case to coerce data type # and check, use out array. result.append(np.mean(a_sorted[tuple(indexer)], axis=axis, out=out)) return np.array(result)
[docs] def get_perspective_matrix(fovy, aspect, z_near, z_far): """ Given a field of view in radians, an aspect ratio, and a near and far plane distance, this routine computes the transformation matrix corresponding to perspective projection using homogeneous coordinates. Parameters ---------- fovy : scalar The angle in degrees of the field of view. aspect : scalar The aspect ratio of width / height for the projection. z_near : scalar The distance of the near plane from the camera. z_far : scalar The distance of the far plane from the camera. Returns ------- persp_matrix : ndarray A new 4x4 2D array. Represents a perspective transformation in homogeneous coordinates. Note that this matrix does not actually perform the projection. After multiplying a 4D vector of the form (x_0, y_0, z_0, 1.0), the point will be transformed to some (x_1, y_1, z_1, w). The final projection is applied by performing a divide by w, that is (x_1/w, y_1/w, z_1/w, w/w). The matrix uses a row-major ordering, rather than the column major ordering typically used by OpenGL. Notes ----- The usage of 4D homogeneous coordinates is for OpenGL and GPU hardware that automatically performs the divide by w operation. See the following for more details about the OpenGL perspective matrices. https://www.tomdalling.com/blog/modern-opengl/explaining-homogenous-coordinates-and-projective-geometry/ http://www.songho.ca/opengl/gl_projectionmatrix.html """ tan_half_fovy = np.tan(np.radians(fovy) / 2) result = np.zeros((4, 4), dtype="float32", order="C") # result[0][0] = 1 / (aspect * tan_half_fovy) # result[1][1] = 1 / tan_half_fovy # result[2][2] = - (z_far + z_near) / (z_far - z_near) # result[3][2] = -1 # result[2][3] = -(2 * z_far * z_near) / (z_far - z_near) f = z_far n = z_near t = tan_half_fovy * n b = -t * aspect r = t * aspect l = -t * aspect result[0][0] = (2 * n) / (r - l) result[2][0] = (r + l) / (r - l) result[1][1] = (2 * n) / (t - b) result[1][2] = (t + b) / (t - b) result[2][2] = -(f + n) / (f - n) result[2][3] = -2 * f * n / (f - n) result[3][2] = -1 return result
[docs] def get_orthographic_matrix(maxr, aspect, z_near, z_far): """ Given a field of view in radians, an aspect ratio, and a near and far plane distance, this routine computes the transformation matrix corresponding to perspective projection using homogeneous coordinates. Parameters ---------- maxr : scalar should be ``max(|x|, |y|)`` aspect : scalar The aspect ratio of width / height for the projection. z_near : scalar The distance of the near plane from the camera. z_far : scalar The distance of the far plane from the camera. Returns ------- persp_matrix : ndarray A new 4x4 2D array. Represents a perspective transformation in homogeneous coordinates. Note that this matrix does not actually perform the projection. After multiplying a 4D vector of the form (x_0, y_0, z_0, 1.0), the point will be transformed to some (x_1, y_1, z_1, w). The final projection is applied by performing a divide by w, that is (x_1/w, y_1/w, z_1/w, w/w). The matrix uses a row-major ordering, rather than the column major ordering typically used by OpenGL. Notes ----- The usage of 4D homogeneous coordinates is for OpenGL and GPU hardware that automatically performs the divide by w operation. See the following for more details about the OpenGL perspective matrices. http://www.scratchapixel.com/lessons/3d-basic-rendering/perspective-and-orthographic-projection-matrix/orthographic-projection-matrix https://www.tomdalling.com/blog/modern-opengl/explaining-homogenous-coordinates-and-projective-geometry/ http://www.songho.ca/opengl/gl_projectionmatrix.html """ r = maxr * aspect t = maxr l = -r b = -t result = np.zeros((4, 4), dtype="float32", order="C") result[0][0] = 2.0 / (r - l) result[1][1] = 2.0 / (t - b) result[2][2] = -2.0 / (z_far - z_near) result[3][3] = 1 result[3][0] = -(r + l) / (r - l) result[3][1] = -(t + b) / (t - b) result[3][2] = -(z_far + z_near) / (z_far - z_near) return result
[docs] def get_lookat_matrix(eye, center, up): """ Given the position of a camera, the point it is looking at, and an up-direction. Computes the lookat matrix that moves all vectors such that the camera is at the origin of the coordinate system, looking down the z-axis. Parameters ---------- eye : array_like The position of the camera. Must be 3D. center : array_like The location that the camera is looking at. Must be 3D. up : array_like The direction that is considered up for the camera. Must be 3D. Returns ------- lookat_matrix : ndarray A new 4x4 2D array in homogeneous coordinates. This matrix moves all vectors in the same way required to move the camera to the origin of the coordinate system, with it pointing down the negative z-axis. """ eye = np.array(eye) center = np.array(center) up = np.array(up) f = (center - eye) / np.linalg.norm(center - eye) s = np.cross(f, up) / np.linalg.norm(np.cross(f, up)) u = np.cross(s, f) result = np.zeros((4, 4), dtype="float32", order="C") result[0][0] = s[0] result[0][1] = s[1] result[0][2] = s[2] result[1][0] = u[0] result[1][1] = u[1] result[1][2] = u[2] result[2][0] = -f[0] result[2][1] = -f[1] result[2][2] = -f[2] result[0][3] = -np.dot(s, eye) result[1][3] = -np.dot(u, eye) result[2][3] = np.dot(f, eye) result[3][3] = 1.0 return result
[docs] def get_translate_matrix(dx, dy, dz): """ Given a movement amount for each coordinate, creates a translation matrix that moves the vector by each amount. Parameters ---------- dx : scalar A translation amount for the x-coordinate dy : scalar A translation amount for the y-coordinate dz : scalar A translation amount for the z-coordinate Returns ------- trans_matrix : ndarray A new 4x4 2D array. Represents a translation by dx, dy and dz in each coordinate respectively. """ result = np.zeros((4, 4), dtype="float32", order="C") result[0][0] = 1.0 result[1][1] = 1.0 result[2][2] = 1.0 result[3][3] = 1.0 result[0][3] = dx result[1][3] = dy result[2][3] = dz return result
[docs] def get_scale_matrix(dx, dy, dz): """ Given a scaling factor for each coordinate, returns a matrix that corresponds to the given scaling amounts. Parameters ---------- dx : scalar A scaling factor for the x-coordinate. dy : scalar A scaling factor for the y-coordinate. dz : scalar A scaling factor for the z-coordinate. Returns ------- scale_matrix : ndarray A new 4x4 2D array. Represents a scaling by dx, dy, and dz in each coordinate respectively. """ result = np.zeros((4, 4), dtype="float32", order="C") result[0][0] = dx result[1][1] = dy result[2][2] = dz result[3][3] = 1 return result
[docs] def get_rotation_matrix(theta, rot_vector): """ Given an angle theta and a 3D vector rot_vector, this routine computes the rotation matrix corresponding to rotating theta radians about rot_vector. Parameters ---------- theta : scalar The angle in radians. rot_vector : array_like The axis of rotation. Must be 3D. Returns ------- rot_matrix : ndarray A new 3x3 2D array. This is the representation of a rotation of theta radians about rot_vector in the simulation box coordinate frame See Also -------- ortho_find Examples -------- >>> a = [0, 1, 0] >>> theta = 0.785398163 # pi/4 >>> rot = mu.get_rotation_matrix(theta, a) >>> rot array([[ 0.70710678, 0. , 0.70710678], [ 0. , 1. , 0. ], [-0.70710678, 0. , 0.70710678]]) >>> np.dot(rot, a) array([ 0., 1., 0.]) # since a is an eigenvector by construction >>> np.dot(rot, [1, 0, 0]) array([ 0.70710678, 0. , -0.70710678]) """ ux = rot_vector[0] uy = rot_vector[1] uz = rot_vector[2] cost = np.cos(theta) sint = np.sin(theta) R = np.array( [ [ cost + ux**2 * (1 - cost), ux * uy * (1 - cost) - uz * sint, ux * uz * (1 - cost) + uy * sint, ], [ uy * ux * (1 - cost) + uz * sint, cost + uy**2 * (1 - cost), uy * uz * (1 - cost) - ux * sint, ], [ uz * ux * (1 - cost) - uy * sint, uz * uy * (1 - cost) + ux * sint, cost + uz**2 * (1 - cost), ], ] ) return R
[docs] def quaternion_mult(q1, q2): """ Multiply two quaternions. The inputs are 4-component numpy arrays in the order [w, x, y, z]. """ w = q1[0] * q2[0] - q1[1] * q2[1] - q1[2] * q2[2] - q1[3] * q2[3] x = q1[0] * q2[1] + q1[1] * q2[0] + q1[2] * q2[3] - q1[3] * q2[2] y = q1[0] * q2[2] + q1[2] * q2[0] + q1[3] * q2[1] - q1[1] * q2[3] z = q1[0] * q2[3] + q1[3] * q2[0] + q1[1] * q2[2] - q1[2] * q2[1] return np.array([w, x, y, z])
[docs] def quaternion_to_rotation_matrix(quaternion): """ This converts a quaternion representation of on orientation to a rotation matrix. The input is a 4-component numpy array in the order [w, x, y, z], and the output is a 3x3 matrix stored as a 2D numpy array. We follow the approach in "3D Math Primer for Graphics and Game Development" by Dunn and Parberry. """ w = quaternion[0] x = quaternion[1] y = quaternion[2] z = quaternion[3] R = np.empty((3, 3), dtype=np.float64) R[0][0] = 1.0 - 2.0 * y**2 - 2.0 * z**2 R[0][1] = 2.0 * x * y + 2.0 * w * z R[0][2] = 2.0 * x * z - 2.0 * w * y R[1][0] = 2.0 * x * y - 2.0 * w * z R[1][1] = 1.0 - 2.0 * x**2 - 2.0 * z**2 R[1][2] = 2.0 * y * z + 2.0 * w * x R[2][0] = 2.0 * x * z + 2.0 * w * y R[2][1] = 2.0 * y * z - 2.0 * w * x R[2][2] = 1.0 - 2.0 * x**2 - 2.0 * y**2 return R
[docs] def rotation_matrix_to_quaternion(rot_matrix): """ Convert a rotation matrix-based representation of an orientation to a quaternion. The input should be a 3x3 rotation matrix, while the output will be a 4-component numpy array. We follow the approach in "3D Math Primer for Graphics and Game Development" by Dunn and Parberry. """ m11 = rot_matrix[0][0] m12 = rot_matrix[0][1] m13 = rot_matrix[0][2] m21 = rot_matrix[1][0] m22 = rot_matrix[1][1] m23 = rot_matrix[1][2] m31 = rot_matrix[2][0] m32 = rot_matrix[2][1] m33 = rot_matrix[2][2] four_w_squared_minus_1 = m11 + m22 + m33 four_x_squared_minus_1 = m11 - m22 - m33 four_y_squared_minus_1 = m22 - m11 - m33 four_z_squared_minus_1 = m33 - m11 - m22 max_index = 0 four_max_squared_minus_1 = four_w_squared_minus_1 if four_x_squared_minus_1 > four_max_squared_minus_1: four_max_squared_minus_1 = four_x_squared_minus_1 max_index = 1 if four_y_squared_minus_1 > four_max_squared_minus_1: four_max_squared_minus_1 = four_y_squared_minus_1 max_index = 2 if four_z_squared_minus_1 > four_max_squared_minus_1: four_max_squared_minus_1 = four_z_squared_minus_1 max_index = 3 max_val = 0.5 * np.sqrt(four_max_squared_minus_1 + 1.0) mult = 0.25 / max_val if max_index == 0: w = max_val x = (m23 - m32) * mult y = (m31 - m13) * mult z = (m12 - m21) * mult elif max_index == 1: x = max_val w = (m23 - m32) * mult y = (m12 + m21) * mult z = (m31 + m13) * mult elif max_index == 2: y = max_val w = (m31 - m13) * mult x = (m12 + m21) * mult z = (m23 + m32) * mult elif max_index == 3: z = max_val w = (m12 - m21) * mult x = (m31 + m13) * mult y = (m23 + m32) * mult return np.array([w, x, y, z])
[docs] def get_sph_r(coords): # The spherical coordinates radius is simply the magnitude of the # coordinate vector. return np.sqrt(np.sum(coords**2, axis=0))
[docs] def resize_vector(vector, vector_array): if len(vector_array.shape) == 4: res_vector = np.resize(vector, (3, 1, 1, 1)) else: res_vector = np.resize(vector, (3, 1)) return res_vector
[docs] def normalize_vector(vector): # this function normalizes # an input vector L2 = np.atleast_1d(np.linalg.norm(vector)) L2[L2 == 0] = 1.0 vector = vector / L2 return vector
[docs] def get_sph_theta(coords, normal): # The angle (theta) with respect to the normal (J), is the arccos # of the dot product of the normal with the normalized coordinate # vector. res_normal = resize_vector(normal, coords) # check if the normal vector is normalized # since arccos requires the vector to be normalised res_normal = normalize_vector(res_normal) tile_shape = [1] + list(coords.shape)[1:] J = np.tile(res_normal, tile_shape) JdotCoords = np.sum(J * coords, axis=0) with np.errstate(invalid="ignore"): ret = np.arccos(JdotCoords / np.sqrt(np.sum(coords**2, axis=0))) ret[np.isnan(ret)] = 0 return ret
[docs] def get_sph_phi(coords, normal): # We have freedom with respect to what axis (xprime) to define # the disk angle. Here I've chosen to use the axis that is # perpendicular to the normal and the y-axis. When normal == # y-hat, then set xprime = z-hat. With this definition, when # normal == z-hat (as is typical), then xprime == x-hat. # # The angle is then given by the arctan of the ratio of the # yprime-component and the xprime-component of the coordinate # vector. normal = normalize_vector(normal) (zprime, xprime, yprime) = ortho_find(normal) res_xprime = resize_vector(xprime, coords) res_yprime = resize_vector(yprime, coords) tile_shape = [1] + list(coords.shape)[1:] Jx = np.tile(res_xprime, tile_shape) Jy = np.tile(res_yprime, tile_shape) Px = np.sum(Jx * coords, axis=0) Py = np.sum(Jy * coords, axis=0) return np.arctan2(Py, Px)
[docs] def get_cyl_r(coords, normal): # The cross product of the normal (J) with a coordinate vector # gives a vector of magnitude equal to the cylindrical radius. res_normal = resize_vector(normal, coords) res_normal = normalize_vector(res_normal) tile_shape = [1] + list(coords.shape)[1:] J = np.tile(res_normal, tile_shape) JcrossCoords = np.cross(J, coords, axisa=0, axisb=0, axisc=0) return np.sqrt(np.sum(JcrossCoords**2, axis=0))
[docs] def get_cyl_z(coords, normal): # The dot product of the normal (J) with the coordinate vector # gives the cylindrical height. res_normal = resize_vector(normal, coords) res_normal = normalize_vector(res_normal) tile_shape = [1] + list(coords.shape)[1:] J = np.tile(res_normal, tile_shape) return np.sum(J * coords, axis=0)
[docs] def get_cyl_theta(coords, normal): # This is identical to the spherical phi component return get_sph_phi(coords, normal)
[docs] def get_cyl_r_component(vectors, theta, normal): # The r of a vector is the vector dotted with rhat normal = normalize_vector(normal) (zprime, xprime, yprime) = ortho_find(normal) res_xprime = resize_vector(xprime, vectors) res_yprime = resize_vector(yprime, vectors) tile_shape = [1] + list(vectors.shape)[1:] Jx = np.tile(res_xprime, tile_shape) Jy = np.tile(res_yprime, tile_shape) rhat = Jx * np.cos(theta) + Jy * np.sin(theta) return np.sum(vectors * rhat, axis=0)
[docs] def get_cyl_theta_component(vectors, theta, normal): # The theta component of a vector is the vector dotted with thetahat normal = normalize_vector(normal) (zprime, xprime, yprime) = ortho_find(normal) res_xprime = resize_vector(xprime, vectors) res_yprime = resize_vector(yprime, vectors) tile_shape = [1] + list(vectors.shape)[1:] Jx = np.tile(res_xprime, tile_shape) Jy = np.tile(res_yprime, tile_shape) thetahat = -Jx * np.sin(theta) + Jy * np.cos(theta) return np.sum(vectors * thetahat, axis=0)
[docs] def get_cyl_z_component(vectors, normal): # The z component of a vector is the vector dotted with zhat normal = normalize_vector(normal) (zprime, xprime, yprime) = ortho_find(normal) res_zprime = resize_vector(zprime, vectors) tile_shape = [1] + list(vectors.shape)[1:] zhat = np.tile(res_zprime, tile_shape) return np.sum(vectors * zhat, axis=0)
[docs] def get_sph_r_component(vectors, theta, phi, normal): # The r component of a vector is the vector dotted with rhat normal = normalize_vector(normal) (zprime, xprime, yprime) = ortho_find(normal) res_xprime = resize_vector(xprime, vectors) res_yprime = resize_vector(yprime, vectors) res_zprime = resize_vector(zprime, vectors) tile_shape = [1] + list(vectors.shape)[1:] Jx, Jy, Jz = ( YTArray(np.tile(rprime, tile_shape), "") for rprime in (res_xprime, res_yprime, res_zprime) ) rhat = ( Jx * np.sin(theta) * np.cos(phi) + Jy * np.sin(theta) * np.sin(phi) + Jz * np.cos(theta) ) return np.sum(vectors * rhat, axis=0)
[docs] def get_sph_phi_component(vectors, phi, normal): # The phi component of a vector is the vector dotted with phihat normal = normalize_vector(normal) (zprime, xprime, yprime) = ortho_find(normal) res_xprime = resize_vector(xprime, vectors) res_yprime = resize_vector(yprime, vectors) tile_shape = [1] + list(vectors.shape)[1:] Jx = YTArray(np.tile(res_xprime, tile_shape), "") Jy = YTArray(np.tile(res_yprime, tile_shape), "") phihat = -Jx * np.sin(phi) + Jy * np.cos(phi) return np.sum(vectors * phihat, axis=0)
[docs] def get_sph_theta_component(vectors, theta, phi, normal): # The theta component of a vector is the vector dotted with thetahat normal = normalize_vector(normal) (zprime, xprime, yprime) = ortho_find(normal) res_xprime = resize_vector(xprime, vectors) res_yprime = resize_vector(yprime, vectors) res_zprime = resize_vector(zprime, vectors) tile_shape = [1] + list(vectors.shape)[1:] Jx, Jy, Jz = ( YTArray(np.tile(rprime, tile_shape), "") for rprime in (res_xprime, res_yprime, res_zprime) ) thetahat = ( Jx * np.cos(theta) * np.cos(phi) + Jy * np.cos(theta) * np.sin(phi) - Jz * np.sin(theta) ) return np.sum(vectors * thetahat, axis=0)
[docs] def compute_stddev_image(buff2, buff): """ This function computes the standard deviation of a weighted projection. It defines the standard deviation as sigma^2 = <v^2>-<v>^2, where the brackets indicate averages (with the weight) and v is the field in question. There is the possibility that if the projection at a particular location is of a constant or a single cell/particle, then <v^2> = <v>^2 and instead of getting zero one gets roundoff error that results in sigma^2 < 0, which is unphysical. We handle this here by performing the subtraction and checking that any and all negative values of sigma^2 can be attributed to roundoff and thus be safely set to zero. We error out if this is not the case. There are ways of computing the standard deviation that are designed to avoid this catastrophic cancellation, but this would require rather substantial and invasive changes to the projection machinery so for the time being it is avoided. Parameters ---------- buff2 : ImageArray The image of the weighted averge of the field squared buff : ImageArray The image of the weighted averge of the field """ buff1 = buff * buff y = buff2 - buff1 close_negs = np.isclose(np.asarray(buff2), np.asarray(buff1))[y < 0.0] if close_negs.all(): y[y < 0.0] = 0.0 else: raise ValueError( "Something went wrong, there are significant negative " "variances in the standard deviation image!" ) return np.sqrt(y)