Related
I have a set of 68 keypoints (size [68, 2]) that I am mapping to gaussian heatmaps. To do this, I have the following function:
def generate_gaussian(t, x, y, sigma=10):
"""
Generates a 2D Gaussian point at location x,y in tensor t.
x should be in range (-1, 1).
sigma is the standard deviation of the generated 2D Gaussian.
"""
h,w = t.shape
# Heatmap pixel per output pixel
mu_x = int(0.5 * (x + 1.) * w)
mu_y = int(0.5 * (y + 1.) * h)
tmp_size = sigma * 3
# Top-left
x1,y1 = int(mu_x - tmp_size), int(mu_y - tmp_size)
# Bottom right
x2, y2 = int(mu_x + tmp_size + 1), int(mu_y + tmp_size + 1)
if x1 >= w or y1 >= h or x2 < 0 or y2 < 0:
return t
size = 2 * tmp_size + 1
tx = np.arange(0, size, 1, np.float32)
ty = tx[:, np.newaxis]
x0 = y0 = size // 2
# The gaussian is not normalized, we want the center value to equal 1
g = torch.tensor(np.exp(- ((tx - x0) ** 2 + (ty - y0) ** 2) / (2 * sigma ** 2)))
# Determine the bounds of the source gaussian
g_x_min, g_x_max = max(0, -x1), min(x2, w) - x1
g_y_min, g_y_max = max(0, -y1), min(y2, h) - y1
# Image range
img_x_min, img_x_max = max(0, x1), min(x2, w)
img_y_min, img_y_max = max(0, y1), min(y2, h)
t[img_y_min:img_y_max, img_x_min:img_x_max] = \
g[g_y_min:g_y_max, g_x_min:g_x_max]
return t
def rescale(a, img_size):
# scale tensor to [-1, 1]
return 2 * a / img_size[0] - 1
My current code uses a for loop to compute the gaussian heatmap for each of the 68 keypoint coordinates, then stacks the resulting tensors to create a [68, H, W] tensor:
x_k1 = [generate_gaussian(torch.zeros(H, W), x, y) for x, y in rescale(kp1.numpy(), frame.shape)]
x_k1 = torch.stack(x_k1, dim=0)
However, this method is super slow. Is there some way that I can do this without a for loop?
Edit:
I tried #Cris Luengo's proposal to compute a 1D Gaussian:
def generate_gaussian1D(t, x, y, sigma=10):
h,w = t.shape
# Heatmap pixel per output pixel
mu_x = int(0.5 * (x + 1.) * w)
mu_y = int(0.5 * (y + 1.) * h)
tmp_size = sigma * 3
# Top-left
x1, y1 = int(mu_x - tmp_size), int(mu_y - tmp_size)
# Bottom right
x2, y2 = int(mu_x + tmp_size + 1), int(mu_y + tmp_size + 1)
if x1 >= w or y1 >= h or x2 < 0 or y2 < 0:
return t
size = 2 * tmp_size + 1
tx = np.arange(0, size, 1, np.float32)
ty = tx[:, np.newaxis]
x0 = y0 = size // 2
g = torch.tensor(np.exp(-np.power(tx - mu_x, 2.) / (2 * np.power(sigma, 2.))))
g = g * g[:, None]
g_x_min, g_x_max = max(0, -x1), min(x2, w) - x1
g_y_min, g_y_max = max(0, -y1), min(y2, h) - y1
img_x_min, img_x_max = max(0, x1), min(x2, w)
img_y_min, img_y_max = max(0, y1), min(y2, h)
t[img_y_min:img_y_max, img_x_min:img_x_max] = \
g[g_y_min:g_y_max, g_x_min:g_x_max]
return t
but my output ends up being an incomplete gaussian.
I'm not sure what I'm doing wrong. Any help would be appreciated.
You generate an NxN array g with a Gaussian centered on its center pixel. N is computed such that it extends by 3*sigma from that center pixel. This is the fastest way to build such an array:
tmp_size = sigma * 3
tx = np.arange(1, tmp_size + 1, 1, np.float32)
g = np.exp(-(tx**2) / (2 * sigma**2))
g = np.concatenate((np.flip(g), [1], g))
g = g * g[:, None]
What we're doing here is compute half a 1D Gaussian. We don't even bother computing the value of the Gaussian for the middle pixel, which we know will be 1. We then build the full 1D Gaussian by flipping our half-Gaussian and concatenating. Finally, the 2D Gaussian is built by the outer product of the 1D Gaussian with itself.
We could shave a bit of extra time by building a quarter of the 2D Gaussian, then concatenating four rotated copies of it. But the difference in computational cost is not very large, and this is much simpler. Note that np.exp is the most expensive operation here by far, so just minimizing how often we call it we significantly reduce the computational cost.
However, the best way to speed up the complete code is to compute the array g only once, rather than anew for each key point. Note how your sigma doesn't change, so all the arrays g that are computed are identical. If you compute it only once, it no longer matters which method you use to compute it, since this will be a minimal portion of the total program anyway.
You could, for example, have a global variable _gaussian to hold your array, and have your function compute it only the first time it is called. Or you could separate your function into two functions, one that constructs this array, and one that copies it into an image, and call them as follows:
g = create_gaussian(sigma=3)
x_k1 = [
copy_gaussian(torch.zeros(H, W), x, y, g)
for x, y in rescale(kp1.numpy(), frame.shape)
]
On the other hand, you're likely best off using existing functionality. For example, DIPlib has a function dip.DrawBandlimitedPoint() [disclosure: I'm an author] that adds a Gaussian blob to an image. Likely you'll find similar functions in other libraries.
I'm trying to create a shallow simulation in 1d (sort of like this), and I've been experimenting with this guide to 2d Shallow Water Equations. However, I don't understand the math all too well, and was wondering if anybody could help me figure out a way to convert this to a 1d system. Any tips or direction would be useful. Thanks.
Here is my code thus far (pretty much all copy/pasted from tutorial):
import numpy
from matplotlib import pyplot
from numba import jit
pyplot.rcParams['font.family'] = 'serif'
pyplot.rcParams['font.size'] = 16
# Update Eta field
#jit(nopython=True) # use Just-In-Time (JIT) Compilation for C-performance
def update_eta_2D(eta, M, N, dx, dy, dt, nx, ny):
# loop over spatial grid
for i in range(1,nx-1):
for j in range(1,ny-1):
# compute spatial derivatives
dMdx = (M[j,i+1] - M[j,i-1]) / (2. * dx)
dNdy = (N[j+1,i] - N[j-1,i]) / (2. * dy)
# update eta field using leap-frog method
eta[j, i] = eta[j, i] - dt * (dMdx + dNdy)
# apply Neumann boundary conditions for eta at all boundaries
eta[0,:] = eta[1,:]
eta[-1,:] = eta[-2,:]
eta[:,0] = eta[:,1]
eta[:,-1] = eta[:,-2]
return eta
# Update M field
#jit(nopython=True) # use Just-In-Time (JIT) Compilation for C-performance
def update_M_2D(eta, M, N, D, g, h, alpha, dx, dy, dt, nx, ny):
# compute argument 1: M**2/D + 0.5 * g * eta**2
arg1 = M**2 / D
# compute argument 2: M * N / D
arg2 = M * N / D
# friction term
fric = g * alpha**2 * M * numpy.sqrt(M**2 + N**2) / D**(7./3.)
# loop over spatial grid
for i in range(1,nx-1):
for j in range(1,ny-1):
# compute spatial derivatives
darg1dx = (arg1[j,i+1] - arg1[j,i-1]) / (2. * dx)
darg2dy = (arg2[j+1,i] - arg2[j-1,i]) / (2. * dy)
detadx = (eta[j,i+1] - eta[j,i-1]) / (2. * dx)
# update M field using leap-frog method
M[j, i] = M[j, i] - dt * (darg1dx + darg2dy + g * D[j,i] * detadx + fric[j,i])
return M
# Update N field
#jit(nopython=True) # use Just-In-Time (JIT) Compilation for C-performance
def update_N_2D(eta, M, N, D, g, h, alpha, dx, dy, dt, nx, ny):
# compute argument 1: M * N / D
arg1 = M * N / D
# compute argument 2: N**2/D + 0.5 * g * eta**2
arg2 = N**2 / D
# friction term
fric = g * alpha**2 * N * numpy.sqrt(M**2 + N**2) / D**(7./3.)
# loop over spatial grid
for i in range(1,nx-1):
for j in range(1,ny-1):
# compute spatial derivatives
darg1dx = (arg1[j,i+1] - arg1[j,i-1]) / (2. * dx)
darg2dy = (arg2[j+1,i] - arg2[j-1,i]) / (2. * dy)
detady = (eta[j+1,i] - eta[j-1,i]) / (2. * dy)
# update N field using leap-frog method
N[j, i] = N[j, i] - dt * (darg1dx + darg2dy + g * D[j,i] * detady + fric[j,i])
return N
# 2D Shallow Water Equation code with JIT optimization
# -------------------------------------------------------
def Shallow_water_2D(eta0, M0, N0, h, g, alpha, nt, dx, dy, dt, X, Y):
"""
Computes and returns the discharge fluxes M, N and wave height eta from
the 2D Shallow water equation using the FTCS finite difference method.
Parameters
----------
eta0 : numpy.ndarray
The initial wave height field as a 2D array of floats.
M0 : numpy.ndarray
The initial discharge flux field in x-direction as a 2D array of floats.
N0 : numpy.ndarray
The initial discharge flux field in y-direction as a 2D array of floats.
h : numpy.ndarray
Bathymetry model as a 2D array of floats.
g : float
gravity acceleration.
alpha : float
Manning's roughness coefficient.
nt : integer
Number fo timesteps.
dx : float
Spatial gridpoint distance in x-direction.
dy : float
Spatial gridpoint distance in y-direction.
dt : float
Time step.
X : numpy.ndarray
x-coordinates as a 2D array of floats.
Y : numpy.ndarray
y-coordinates as a 2D array of floats.
Returns
-------
eta : numpy.ndarray
The final wave height field as a 2D array of floats.
M : numpy.ndarray
The final discharge flux field in x-direction as a 2D array of floats.
N : numpy.ndarray
The final discharge flux field in y-direction as a 2D array of floats.
"""
# Copy fields
eta = eta0.copy()
M = M0.copy()
N = N0.copy()
# Compute total thickness of water column D
D = eta + h
# Estimate number of grid points in x- and y-direction
ny, nx = eta.shape
# Define the locations along a gridline.
x = numpy.linspace(0, nx*dx, num=nx)
y = numpy.linspace(0, ny*dy, num=ny)
# Plot the initial wave height fields eta and bathymetry model
fig = pyplot.figure(figsize=(10., 6.))
cmap = 'seismic'
pyplot.tight_layout()
extent = [numpy.min(x), numpy.max(x),numpy.min(y), numpy.max(y)]
# Plot bathymetry model
topo = pyplot.imshow(numpy.flipud(-h), cmap=pyplot.cm.gray, interpolation='nearest', extent=extent)
# Plot wave height field at current time step
im = pyplot.imshow(numpy.flipud(eta), extent=extent, interpolation='spline36', cmap=cmap, alpha=.75, vmin = -0.4, vmax=0.4)
pyplot.xlabel('x [m]')
pyplot.ylabel('y [m]')
cbar = pyplot.colorbar(im)
cbar1 = pyplot.colorbar(topo)
pyplot.gca().invert_yaxis()
cbar.set_label(r'$\eta$ [m]')
cbar1.set_label(r'$-h$ [m]')
# activate interactive plot
pyplot.ion()
pyplot.show(block=False)
# write wave height field and bathymetry snapshots every nsnap time steps to image file
nsnap = 50
snap_count = 0
# Loop over timesteps
for n in range(nt):
# 1. Update Eta field
# -------------------
eta = update_eta_2D(eta, M, N, dx, dy, dt, nx, ny)
# 2. Update M field
# -----------------
M = update_M_2D(eta, M, N, D, g, h, alpha, dx, dy, dt, nx, ny)
# 3. Update N field
# -----------------
N = update_N_2D(eta, M, N, D, g, h, alpha, dx, dy, dt, nx, ny)
# 4. Compute total water column D
# -------------------------------
D = eta + h
# update wave height field eta
if (n % nsnap) == 0:
im.set_data(eta)
fig.canvas.draw()
# write snapshots to Tiff files
name_snap = "image_out/Shallow_water_2D_" + "%0.*f" %(0,numpy.fix(snap_count+1000)) + ".tiff"
pyplot.savefig(name_snap, format='tiff', bbox_inches='tight', dpi=125)
snap_count += 1
return eta, M, N
Lx = 100.0 # width of the mantle in the x direction []
Ly = 100.0 # thickness of the mantle in the y direction []
nx = 401 # number of points in the x direction
ny = 401 # number of points in the y direction
dx = Lx / (nx - 1) # grid spacing in the x direction []
dy = Ly / (ny - 1) # grid spacing in the y direction []
# Define the locations along a gridline.
x = numpy.linspace(0.0, Lx, num=nx)
y = numpy.linspace(0.0, Ly, num=ny)
# Define initial eta, M, N
X, Y = numpy.meshgrid(x,y) # coordinates X,Y required to define eta, h, M, N
# Define constant ocean depth profile h = 50 m
h = 50 * numpy.ones_like(X)
# Define initial eta Gaussian distribution [m]
eta0 = 0.5 * numpy.exp(-((X-50)**2/10)-((Y-50)**2/10))
# Define initial M and N
M0 = 100. * eta0
N0 = 0. * M0
# define some constants
g = 9.81 # gravity acceleration [m/s^2]
alpha = 0.025 # friction coefficient for natural channels in good condition
# Maximum wave propagation time [s]
Tmax = 6.
dt = 1/4500.
nt = (int)(Tmax/dt)
# Compute eta, M, N fields
eta, M, N = Shallow_water_2D(eta0, M0, N0, h, g, alpha, nt, dx, dy, dt, X, Y)
If all you want is the 1D version of the model:
I have a simple 2D ray-casting routine that gets terribly slow as soon as the number of obstacles increases.
This routine is made up of:
2 for loops (outer loop iterates over each ray/direction, then inner loop iterates over each line obstacle)
multiple if statements (check if a value is > or < than another value or if an array is empty)
Question: How can I condense all these operations into 1 single block of vectorized instructions using Numpy ?
More specifically, I am facing 2 issues:
I have managed to vectorize the inner loop (intersection between a ray and each obstacle) but I am unable to run this operation for all rays at once.
The only workaround I found to deal with the if statements is to use masked arrays. Something tells me it is not the proper way to handle these statements in this case (it seems clumsy, cumbersome and unpythonic)
Original code:
from math import radians, cos, sin
import matplotlib.pyplot as plt
import numpy as np
N = 10 # dimensions of canvas (NxN)
sides = np.array([[0, N, 0, 0], [0, N, N, N], [0, 0, 0, N], [N, N, 0, N]])
edges = np.random.rand(5, 4) * N # coordinates of 5 random segments (x1, x2, y1, y2)
edges = np.concatenate((edges, sides))
center = np.array([N/2, N/2]) # coordinates of center point
directions = np.array([(cos(radians(a)), sin(radians(a))) for a in range(0, 360, 10)]) # vectors pointing in all directions
intersections = []
# for each direction
for d in directions:
min_dist = float('inf')
# for each edge
for e in edges:
p1x, p1y = e[0], e[2]
p2x, p2y = e[1], e[3]
p3x, p3y = center
p4x, p4y = center + d
# find intersection point
den = (p1x - p2x) * (p3y - p4y) - (p1y - p2y) * (p3x - p4x)
if den:
t = ((p1x - p3x) * (p3y - p4y) - (p1y - p3y) * (p3x - p4x)) / den
u = -((p1x - p2x) * (p1y - p3y) - (p1y - p2y) * (p1x - p3x)) / den
# if any:
if t > 0 and t < 1 and u > 0:
sx = p1x + t * (p2x - p1x)
sy = p1y + t * (p2y - p1y)
isec = np.array([sx, sy])
dist = np.linalg.norm(isec-center)
# make sure to select the nearest one (from center)
if dist < min_dist:
min_dist = dist
nearest = isec
# store nearest interesection point for each ray
intersections.append(nearest)
# Render
plt.axis('off')
for x, y in zip(edges[:,:2], edges[:,2:]):
plt.plot(x, y)
for isec in np.array(intersections):
plt.plot((center[0], isec[0]), (center[1], isec[1]), '--', color="#aaaaaa", linewidth=.8)
Vectorized version (attempt):
from math import radians, cos, sin
import matplotlib.pyplot as plt
from scipy import spatial
import numpy as np
N = 10 # dimensions of canvas (NxN)
sides = np.array([[0, N, 0, 0], [0, N, N, N], [0, 0, 0, N], [N, N, 0, N]])
edges = np.random.rand(5, 4) * N # coordinates of 5 random segments (x1, x2, y1, y2)
edges = np.concatenate((edges, sides))
center = np.array([N/2, N/2]) # coordinates of center point
directions = np.array([(cos(radians(a)), sin(radians(a))) for a in range(0, 360, 10)]) # vectors pointing in all directions
intersections = []
# Render edges
plt.axis('off')
for x, y in zip(edges[:,:2], edges[:,2:]):
plt.plot(x, y)
# for each direction
for d in directions:
p1x, p1y = edges[:,0], edges[:,2]
p2x, p2y = edges[:,1], edges[:,3]
p3x, p3y = center
p4x, p4y = center + d
# denominator
den = (p1x - p2x) * (p3y - p4y) - (p1y - p2y) * (p3x - p4x)
# first 'if' statement -> if den > 0
mask = den > 0
den = den[mask]
p1x = p1x[mask]
p1y = p1y[mask]
p2x = p2x[mask]
p2y = p2y[mask]
t = ((p1x - p3x) * (p3y - p4y) - (p1y - p3y) * (p3x - p4x)) / den
u = -((p1x - p2x) * (p1y - p3y) - (p1y - p2y) * (p1x - p3x)) / den
# second 'if' statement -> if (t>0) & (t<1) & (u>0)
mask2 = (t > 0) & (t < 1) & (u > 0)
t = t[mask2]
p1x = p1x[mask2]
p1y = p1y[mask2]
p2x = p2x[mask2]
p2y = p2y[mask2]
# x, y coordinates of all intersection points in the current direction
sx = p1x + t * (p2x - p1x)
sy = p1y + t * (p2y - p1y)
pts = np.c_[sx, sy]
# if any:
if pts.size > 0:
# find nearest intersection point
tree = spatial.KDTree(pts)
nearest = pts[tree.query(center)[1]]
# Render
plt.plot((center[0], nearest[0]), (center[1], nearest[1]), '--', color="#aaaaaa", linewidth=.8)
Reformulation of the problem – Finding the intersection between a line segment and a line ray
Let q and q2 be the endpoints of a segment (obstacle). For convenience let's define a class to represent points and vectors in the plane. In addition to the usual operations, a vector multiplication is defined by u × v = u.x * v.y - u.y * v.x.
Caution: here Coord(2, 1) * 3 returns Coord(6, 3) while Coord(2, 1) * Coord(-1, 4) outputs 9. To avoid this confusion it might have been possible to restrict * to the scalar multiplication and use ^ via __xor__ for the vector multiplication.
class Coord:
def __init__(self, x, y):
self.x = x
self.y = y
#property
def radius(self):
return np.sqrt(self.x ** 2 + self.y ** 2)
def _cross_product(self, other):
assert isinstance(other, Coord)
return self.x * other.y - self.y * other.x
def __mul__(self, other):
if isinstance(other, Coord):
# 2D "cross"-product
return self._cross_product(other)
elif isinstance(other, int) or isinstance(other, float):
# scalar multiplication
return Coord(self.x * other, self.y * other)
def __rmul__(self, other):
return self * other
def __sub__(self, other):
return Coord(self.x - other.x, self.y - other.y)
def __add__(self, other):
return Coord(self.x + other.x, self.y + other.y)
def __repr__(self):
return f"Coord({self.x}, {self.y})"
Now, I find it easier to handle a ray in polar coordinates: For a given angle theta (direction) the goal is to determine if it intersects the segment, and if so determine the corresponding radius. Here is a function to find that. See here for an explanation of why and how. I tried to use the same variable names as in the previous link.
def find_intersect_btw_ray_and_sgmt(q, q2, theta):
"""
Args:
q (Coord): first endpoint of the segment
q2 (Coord): second endpoint of the segment
theta (float): angle of the ray
Returns:
(float): np.inf if the ray does not intersect the segment,
the distance from the origin of the intersection otherwise
"""
assert isinstance(q, Coord) and isinstance(q2, Coord)
s = q2 - q
r = Coord(np.cos(theta), np.sin(theta))
cross = r * s # 2d cross-product
t_num = q * s
u_num = q * r
## the intersection point is roughly at a distance t_num / cross
## from the origin. But some cases must be checked beforehand.
## (1) the segment [PQ2] is aligned with the ray
if np.isclose(cross, 0) and np.isclose(u_num, 0):
return min(q.radius, q2.radius)
## (2) the segment [PQ2] is parallel with the ray
elif np.isclose(cross, 0):
return np.inf
t, u = t_num / cross, u_num / cross
## There is actually an intersection point
if t >= 0 and 0 <= u <= 1:
return t
## (3) No intersection point
return np.inf
For instance find_intersect_btw_ray_and_sgmt(Coord(1, 2), Coord(-1, 2), np.pi / 2) should returns 2.
Note that here for simplicity, I only considered the case where the origin of the rays is at Coord(0, 0). This can be easily extended to the general case by setting t_num = (q - origin) * s and u_num = (q - origin) * r.
Let's vectorize it!
What is very interesting here is that the operations defined in the Coord class also apply to cases where x and y are numpy arrays! Hence applying any defined operation on Coord(np.array([1, 2, 0]), np.array([2, -1, 3])) amounts applying it elementwise to the points (1, 2), (2, -1) and (0, 3). The operations of Coord are therefore already vectorized. The constructor can be modified into:
def __init__(self, x, y):
x, y = np.array(x), np.array(y)
assert x.shape == y.shape
self.x, self.y = x, y
self.shape = x.shape
Now, we would like the function find_intersect_btw_ray_and_sgmt to be able to handle the case where the parameters q and q2contains sequences of endpoints. Before the sanity checks, all the operations are working properly since, as we have mentioned, they are already vectorized. As you mentionned the conditional statements can be "vectorized" using masks. Here is what I propose:
def find_intersect_btw_ray_and_sgmts(q, q2, theta):
assert isinstance(q, Coord) and isinstance(q2, Coord)
assert q.shape == q2.shape
EPS = 1e-14
s = q2 - q
r = Coord(np.cos(theta), np.sin(theta))
cross = r * s
cross_sign = np.sign(cross)
cross = cross * cross_sign
t_num = (q * s) * cross_sign
u_num = (q * r) * cross_sign
radii = np.zeros_like(t_num)
mask = ~np.isclose(cross, 0) & (t_num >= -EPS) & (-EPS <= u_num) & (u_num <= cross + EPS)
radii[~mask] = np.inf # no intersection
radii[mask] = t_num[mask] / cross[mask] # intersection
return radii
Note that cross, t_num and u_num are multiplied by the sign of cross to ensure that the division by cross keeps the sign of the dividends. Hence conditions of the form ((t_num >= 0) & (cross >= 0)) | ((t_num <= 0) & (cross <= 0)) can be replaced by (t_num >= 0).
For simplicity, we omitted the case (1) where the radius and the segment were aligned ((cross == 0) & (u_num == 0)). This could be incorporated by carefully adding a second mask.
For a given value of theta, we are able to determine if the corresponing ray intersects with several segments at once.
## Some useful functions
def polar_to_cartesian(r, theta):
return Coord(r * np.cos(theta), r * np.sin(theta))
def plot_segments(p, q, *args, **kwargs):
plt.plot([p.x, q.x], [p.y, q.y], *args, **kwargs)
def plot_rays(radii, thetas, *args, **kwargs):
endpoints = polar_to_cartesian(radii, thetas)
n = endpoints.shape
origin = Coord(np.zeros(n), np.zeros(n))
plot_segments(origin, endpoints, *args, **kwargs)
## Data generation
M = 5 # size of the canvas
N = 10 # number of segments
K = 16 # number of rays
q = Coord(*np.random.uniform(-M/2, M/2, size=(2, N)))
p = q + Coord(*np.random.uniform(-M/2, M/2, size=(2, N)))
thetas = np.linspace(0, 2 * np.pi, K, endpoint=False)
## For each ray, find the minimal distance of intersection
## with all segments
plt.figure(figsize=(5, 5))
plot_segments(p, q, "royalblue", marker=".")
for theta in thetas:
radii = find_intersect_btw_ray_and_sgmts(p, q, theta)
radius = np.min(radii)
if not np.isinf(radius):
plot_rays(radius, theta, color="orange")
else:
plot_rays(2*M, theta, ':', c='orange')
plt.plot(0, 0, 'kx')
plt.xlim(-M, M)
plt.ylim(-M, M)
And that's not all! Thanks to the broadcasting of python, it is possible to avoid iteration on theta values. For example, recall that np.array([1, 2, 3]) * np.array([[1], [2], [3], [4]]) produces a matrix of size 4 × 3 of the pairwise products. In the same way Coord([[5],[7]], [[5],[1]]) * Coord([2, 4, 6], [-2, 4, 0]) outputs a 2 × 3 matrix containing all the pairwise cross product between vectors (5, 5), (7, 1) and (2, -2), (4, 4), (6, 0).
Finally, the intersections can be determined in the following way:
radii_all = find_intersect_btw_ray_and_sgmts(p, q, np.vstack(thetas))
# p and q have a shape of (N,) and np.vstack(thetas) of (K, 1)
# this radii_all have a shape of (K, N)
# radii_all[k, n] contains the distance from the origin of the intersection
# between k-th ray and n-th segment (or np.inf if there is no intersection point)
radii = np.min(radii_all, axis=1)
# radii[k] contains the distance from the origin of the closest intersection
# between k-th ray and all segments
do_intersect = ~np.isinf(radii)
plot_rays(radii[do_intersect], thetas[do_intersect], color="orange")
plot_rays(2*M, thetas[~do_intersect], ":", color="orange")
I'm able to use numpy.polynomial to fit terms to 1D polynomials like f(x) = 1 + x + x^2. How can I fit multidimensional polynomials, like f(x,y) = 1 + x + x^2 + y + yx + y x^2 + y^2 + y^2 x + y^2 x^2? It looks like numpy doesn't support multidimensional polynomials at all: is that the case? In my real application, I have 5 dimensions of input and I am interested in hermite polynomials. It looks like the polynomials in scipy.special are also only available for one dimension of inputs.
# One dimension of data can be fit
x = np.random.random(100)
y = np.sin(x)
params = np.polynomial.polynomial.polyfit(x, y, 6)
np.polynomial.polynomial.polyval([0, .2, .5, 1.5], params)
array([ -5.01799432e-08, 1.98669317e-01, 4.79425535e-01,
9.97606096e-01])
# When I try two dimensions, it fails.
x = np.random.random((100, 2))
y = np.sin(5 * x[:,0]) + .4 * np.sin(x[:,1])
params = np.polynomial.polynomial.polyvander2d(x, y, [6, 6])
---------------------------------------------------------------------------
ValueError Traceback (most recent call last)
<ipython-input-13-5409f9a3e632> in <module>()
----> 1 params = np.polynomial.polynomial.polyvander2d(x, y, [6, 6])
/usr/local/lib/python2.7/site-packages/numpy/polynomial/polynomial.pyc in polyvander2d(x, y, deg)
1201 raise ValueError("degrees must be non-negative integers")
1202 degx, degy = ideg
-> 1203 x, y = np.array((x, y), copy=0) + 0.0
1204
1205 vx = polyvander(x, degx)
ValueError: could not broadcast input array from shape (100,2) into shape (100)
I got annoyed that there is no simple function for a 2d polynomial fit of any number of degrees so I made my own. Like the other answers it uses numpy lstsq to find the best coefficients.
import numpy as np
from scipy.linalg import lstsq
from scipy.special import binom
import matplotlib.pyplot as plt
from mpl_toolkits.mplot3d import Axes3D
def _get_coeff_idx(coeff):
idx = np.indices(coeff.shape)
idx = idx.T.swapaxes(0, 1).reshape((-1, 2))
return idx
def _scale(x, y):
# Normalize x and y to avoid huge numbers
# Mean 0, Variation 1
offset_x, offset_y = np.mean(x), np.mean(y)
norm_x, norm_y = np.std(x), np.std(y)
x = (x - offset_x) / norm_x
y = (y - offset_y) / norm_y
return x, y, (norm_x, norm_y), (offset_x, offset_y)
def _unscale(x, y, norm, offset):
x = x * norm[0] + offset[0]
y = y * norm[1] + offset[1]
return x, y
def polyvander2d(x, y, degree):
A = np.polynomial.polynomial.polyvander2d(x, y, degree)
return A
def polyscale2d(coeff, scale_x, scale_y, copy=True):
if copy:
coeff = np.copy(coeff)
idx = _get_coeff_idx(coeff)
for k, (i, j) in enumerate(idx):
coeff[i, j] /= scale_x ** i * scale_y ** j
return coeff
def polyshift2d(coeff, offset_x, offset_y, copy=True):
if copy:
coeff = np.copy(coeff)
idx = _get_coeff_idx(coeff)
# Copy coeff because it changes during the loop
coeff2 = np.copy(coeff)
for k, m in idx:
not_the_same = ~((idx[:, 0] == k) & (idx[:, 1] == m))
above = (idx[:, 0] >= k) & (idx[:, 1] >= m) & not_the_same
for i, j in idx[above]:
b = binom(i, k) * binom(j, m)
sign = (-1) ** ((i - k) + (j - m))
offset = offset_x ** (i - k) * offset_y ** (j - m)
coeff[k, m] += sign * b * coeff2[i, j] * offset
return coeff
def plot2d(x, y, z, coeff):
# regular grid covering the domain of the data
if x.size > 500:
choice = np.random.choice(x.size, size=500, replace=False)
else:
choice = slice(None, None, None)
x, y, z = x[choice], y[choice], z[choice]
X, Y = np.meshgrid(
np.linspace(np.min(x), np.max(x), 20), np.linspace(np.min(y), np.max(y), 20)
)
Z = np.polynomial.polynomial.polyval2d(X, Y, coeff)
fig = plt.figure()
ax = fig.gca(projection="3d")
ax.plot_surface(X, Y, Z, rstride=1, cstride=1, alpha=0.2)
ax.scatter(x, y, z, c="r", s=50)
plt.xlabel("X")
plt.ylabel("Y")
ax.set_zlabel("Z")
plt.show()
def polyfit2d(x, y, z, degree=1, max_degree=None, scale=True, plot=False):
"""A simple 2D polynomial fit to data x, y, z
The polynomial can be evaluated with numpy.polynomial.polynomial.polyval2d
Parameters
----------
x : array[n]
x coordinates
y : array[n]
y coordinates
z : array[n]
data values
degree : {int, 2-tuple}, optional
degree of the polynomial fit in x and y direction (default: 1)
max_degree : {int, None}, optional
if given the maximum combined degree of the coefficients is limited to this value
scale : bool, optional
Wether to scale the input arrays x and y to mean 0 and variance 1, to avoid numerical overflows.
Especially useful at higher degrees. (default: True)
plot : bool, optional
wether to plot the fitted surface and data (slow) (default: False)
Returns
-------
coeff : array[degree+1, degree+1]
the polynomial coefficients in numpy 2d format, i.e. coeff[i, j] for x**i * y**j
"""
# Flatten input
x = np.asarray(x).ravel()
y = np.asarray(y).ravel()
z = np.asarray(z).ravel()
# Remove masked values
mask = ~(np.ma.getmask(z) | np.ma.getmask(x) | np.ma.getmask(y))
x, y, z = x[mask].ravel(), y[mask].ravel(), z[mask].ravel()
# Scale coordinates to smaller values to avoid numerical problems at larger degrees
if scale:
x, y, norm, offset = _scale(x, y)
if np.isscalar(degree):
degree = (int(degree), int(degree))
degree = [int(degree[0]), int(degree[1])]
coeff = np.zeros((degree[0] + 1, degree[1] + 1))
idx = _get_coeff_idx(coeff)
# Calculate elements 1, x, y, x*y, x**2, y**2, ...
A = polyvander2d(x, y, degree)
# We only want the combinations with maximum order COMBINED power
if max_degree is not None:
mask = idx[:, 0] + idx[:, 1] <= int(max_degree)
idx = idx[mask]
A = A[:, mask]
# Do the actual least squares fit
C, *_ = lstsq(A, z)
# Reorder coefficients into numpy compatible 2d array
for k, (i, j) in enumerate(idx):
coeff[i, j] = C[k]
# Reverse the scaling
if scale:
coeff = polyscale2d(coeff, *norm, copy=False)
coeff = polyshift2d(coeff, *offset, copy=False)
if plot:
if scale:
x, y = _unscale(x, y, norm, offset)
plot2d(x, y, z, coeff)
return coeff
if __name__ == "__main__":
n = 100
x, y = np.meshgrid(np.arange(n), np.arange(n))
z = x ** 2 + y ** 2
c = polyfit2d(x, y, z, degree=2, plot=True)
print(c)
It doesn't look like polyfit supports fitting multivariate polynomials, but you can do it by hand, with linalg.lstsq. The steps are as follows:
Gather the degrees of monomials x**i * y**j you wish to use in the model. Think carefully about it: your current model already has 9 parameters, if you are going to push to 5 variables then with the current approach you'll end up with 3**5 = 243 parameters, a sure road to overfitting. Maybe limit to the monomials of __total_ degree at most 2 or three...
Plug the x-points into each monomial; this gives a 1D array. Stack all such arrays as columns of a matrix.
Solve a linear system with aforementioned matrix and with the right-hand side being the target values (I call them z because y is confusing when you also use x, y for two variables).
Here it is:
import numpy as np
x = np.random.random((100, 2))
z = np.sin(5 * x[:,0]) + .4 * np.sin(x[:,1])
degrees = [(i, j) for i in range(3) for j in range(3)] # list of monomials x**i * y**j to use
matrix = np.stack([np.prod(x**d, axis=1) for d in degrees], axis=-1) # stack monomials like columns
coeff = np.linalg.lstsq(matrix, z)[0] # lstsq returns some additional info we ignore
print("Coefficients", coeff) # in the same order as the monomials listed in "degrees"
fit = np.dot(matrix, coeff)
print("Fitted values", fit)
print("Original values", y)
I believe you have misunderstood what polyvander2d does and how it should be used. polyvander2d() returns the pseudo-Vandermonde matrix of degrees deg and sample points (x, y).
Here, y is not the value(s) of the polynomial at point(s) x but rather it is the y-coordinate of the point(s) and x is the x-coordinate. Roughly speaking, the returned array is a set of combinations of (x**i) * (y**j) and x and y are essentially 2D "mesh-grids". Therefore, both x and y must have identical shapes.
Your x and y, however, arrays have different shapes:
>>> x.shape
(100, 2)
>>> y.shape
(100,)
I do not believe numpy has a 5D-polyvander of the form polyvander5D(x, y, z, v, w, deg). Notice, all the variables here are coordinates and not the values of the polynomial p=p(x,y,z,v,w). You, however, seem to be using y (in the 2D case) as f.
It appears that numpy does not have 2D or higher equivalents for the polyfit() function. If your intention is to find the coefficients of the best-fitting polynomial in higher-dimensions, I would suggest that you generalize the approach described here: Equivalent of `polyfit` for a 2D polynomial in Python
The option isn't there because nobody wants to do that. Combine the polynomials linearly (f(x,y) = 1 + x + y + x^2 + y^2) and solve the system of equations yourself.
I'd like to initialize a numpy array to represent a two-dimensional vector field on a 100 x 100 grid of points defined by:
import numpy as np
dx = dy = 0.1
nx = ny = 100
x, y = np.meshgrid(np.arange(0,nx*dx,dx), np.arange(0,ny*dy,dy))
The field is a constant-speed circulation about the point cx,cy and I can initialize it OK with regular Python loops:
v = np.empty((nx, ny, 2))
cx, cy = 5, 5
s = 2
for i in range(nx):
for j in range(ny):
rx, ry = i*dx - cx, j*dy - cy
r = np.hypot(rx, ry)
if r == 0:
v[i,j] = 0,0
continue
# (-ry/r, rx/r): the unit vector tangent to the circle centred at (cx,cy), radius r
v[i,j] = (s * -ry/r, s * rx/r)
But when I'm having trouble vectorizing with numpy. The closest I've got is
v = np.array([s * -(y-cy) / np.hypot(x-cx, y-cy), s * (x-cx) / np.hypot(x-cx, y-cy)])
v = np.rollaxis(v, 1, 0)
v = np.rollaxis(v, 2, 1)
v[np.isinf(v)] = 0
But this isn't equivalent and doesn't give the right answer. What is the correct way to initialize a vector field using numpy?
EDIT: OK - now I'm confused following the suggestion below, I try:
vx = s * -(y-cy) / np.hypot(x-cx, y-cy)
vy = s * (x-cx) / np.hypot(x-cx, y-cy)
v = np.dstack((vx, vy))
v[np.isnan(v)] = 0
but get a completely different array...
From your initial setup:
import numpy as np
dx = dy = 0.1
nx = ny = 100
x, y = np.meshgrid(np.arange(0, nx * dx, dx),
np.arange(0, ny * dy, dy))
cx = cy = 5
s = 2
You could compute v like this:
rx, ry = y - cx, x - cy
r = np.hypot(rx, ry)
v2 = s * np.dstack((-ry, rx)) / r[..., None]
v2[np.isnan(v2)] = 0
If you're feeling really fancy, you could create yx as a 3D array, and broadcast all of the operations over it:
# we make these [2,] arrays to broadcast over the last output dimension
c = np.array([5, 5])
s = np.array([-2, 2])
# this creates a [100, 100, 2] mesh, where the last dimension corresponds
# to (y, x)
yx = np.mgrid[0:nx * dx:dx, 0:ny * dy:dy].T
yxdiff = yx - c[None, None, :]
r = np.hypot(yxdiff[..., 0], yxdiff[..., 1])[..., None]
v3 = s[None, None, :] * yxdiff / r
v3[np.isnan(v3)] = 0
Check that these both give the same answer as your original code:
print np.all(v == v2), np.all(v == v3)
# True, True
Edit
Why rx, ry = y - cx, x - cy rather than rx, ry = x - cx, y - cy? I agree it's very counterintuitive - the only reason I decided to do it that way was to match the output of your original code.
The issue is that in your grids, consecutive x values are actually found in consecutive columns of x, and consecutive y values are found in consecutive rows of y, i.e. x[:, j] is the j th x-value and y[i, :] is the i th y-value. However, in your inner loop, you are multiplying dx by i, which is your row index, and dy by j, which is your column index. You're therefore flipping the x and y dimensions of your output.