Fast weighted euclidean distance between points in arrays - python

I need to efficiently calculate the euclidean weighted distances for every x,y point in a given array to every other x,y point in another array. This is the code I have which works as expected:
import numpy as np
import random
def rand_data(integ):
'''
Function that generates 'integ' random values between [0.,1.)
'''
rand_dat = [random.random() for _ in range(integ)]
return rand_dat
def weighted_dist(indx, x_coo, y_coo):
'''
Function that calculates *weighted* euclidean distances.
'''
dist_point_list = []
# Iterate through every point in array_2.
for indx2, x_coo2 in enumerate(array_2[0]):
y_coo2 = array_2[1][indx2]
# Weighted distance in x.
x_dist_weight = (x_coo-x_coo2)/w_data[0][indx]
# Weighted distance in y.
y_dist_weight = (y_coo-y_coo2)/w_data[1][indx]
# Weighted distance between point from array_1 passed and this point
# from array_2.
dist = np.sqrt(x_dist_weight**2 + y_dist_weight**2)
# Append weighted distance value to list.
dist_point_list.append(round(dist, 8))
return dist_point_list
# Generate random x,y data points.
array_1 = np.array([rand_data(10), rand_data(10)], dtype=float)
# Generate weights for each x,y coord for points in array_1.
w_data = np.array([rand_data(10), rand_data(10)], dtype=float)
# Generate second larger array.
array_2 = np.array([rand_data(100), rand_data(100)], dtype=float)
# Obtain *weighted* distances for every point in array_1 to every point in array_2.
dist = []
# Iterate through every point in array_1.
for indx, x_coo in enumerate(array_1[0]):
y_coo = array_1[1][indx]
# Call function to get weighted distances for this point to every point in
# array_2.
dist.append(weighted_dist(indx, x_coo, y_coo))
The final list dist holds as many sub-lists as points are in the first array with as many elements in each as points are in the second one (the weighted distances).
I'd like to know if there's a way to make this code more efficient, perhaps using the cdist function, because this process becomes quite expensive when the arrays have lots of elements (which in my case they have) and when I have to check the distances for lots of arrays (which I also have)

#Evan and #Martinis Group are on the right track - to expand on Evan's answer, here's a function that uses broadcasting to quickly calculate the n-dimensional weighted euclidean distance without Python loops:
import numpy as np
def fast_wdist(A, B, W):
"""
Compute the weighted euclidean distance between two arrays of points:
D{i,j} =
sqrt( ((A{0,i}-B{0,j})/W{0,i})^2 + ... + ((A{k,i}-B{k,j})/W{k,i})^2 )
inputs:
A is an (k, m) array of coordinates
B is an (k, n) array of coordinates
W is an (k, m) array of weights
returns:
D is an (m, n) array of weighted euclidean distances
"""
# compute the differences and apply the weights in one go using
# broadcasting jujitsu. the result is (n, k, m)
wdiff = (A[np.newaxis,...] - B[np.newaxis,...].T) / W[np.newaxis,...]
# square and sum over the second axis, take the sqrt and transpose. the
# result is an (m, n) array of weighted euclidean distances
D = np.sqrt((wdiff*wdiff).sum(1)).T
return D
To check that this works OK, we'll compare it to a slower version that uses nested Python loops:
def slow_wdist(A, B, W):
k,m = A.shape
_,n = B.shape
D = np.zeros((m, n))
for ii in xrange(m):
for jj in xrange(n):
wdiff = (A[:,ii] - B[:,jj]) / W[:,ii]
D[ii,jj] = np.sqrt((wdiff**2).sum())
return D
First, let's make sure that the two functions give the same answer:
# make some random points and weights
def setup(k=2, m=100, n=300):
return np.random.randn(k,m), np.random.randn(k,n),np.random.randn(k,m)
a, b, w = setup()
d0 = slow_wdist(a, b, w)
d1 = fast_wdist(a, b, w)
print np.allclose(d0, d1)
# True
Needless to say, the version that uses broadcasting rather than Python loops is several orders of magnitude faster:
%%timeit a, b, w = setup()
slow_wdist(a, b, w)
# 1 loops, best of 3: 647 ms per loop
%%timeit a, b, w = setup()
fast_wdist(a, b, w)
# 1000 loops, best of 3: 620 us per loop

You could use cdist if you don't need weighted distances. If you need weighted distances and performance, create an array of the appropriate output size, and use either an automated accelerator like Numba or Parakeet, or hand-tune the code with Cython.

You can avoid looping by using code that looks like the following:
def compute_distances(A, B, W):
Ax = A[:,0].reshape(1, A.shape[0])
Bx = B[:,0].reshape(A.shape[0], 1)
dx = Bx-Ax
# Same for dy
dist = np.sqrt(dx**2 + dy**2) * W
return dist
That will run a lot faster in python that anything that loops as long as you have enough memory for the arrays.

You could try removing the square root, since if a>b, it follows that a squared > b squared... and computers are REALLY slow at square roots normally.

Related

Algorithm to pixelate positions and fluxes

What would you do if you had n particles on a plane (with positions (x_n,y_n)), with a certain flux flux_n, and you have to pixelate these particles, so you have to go from (x,y) to (pixel_i, pixel_j) space and you have to sum up the flux of the m particles which fall in to every single pixel? Any suggestions? Thank you!
The are several ways with which you can solve your problem.
Assumptions: your positions have been stored into two numpy array of shape (N, ), i.e. the position x_n (or y_n) for n in [0, N), let's call them x and y. The flux is stored into a numpy array with the same shape, fluxes.
1 - INTENSIVE CASE
Create something that looks like a grid:
#get minimums and maximums position
mins = int(x.min()), int(y.min())
maxs = int(x.max()), int(y.max())
#actually you can also add and subtract 1 or more unit
#in order to have a grid larger than the x, y extremes
#something like mins-=epsilon and maxs += epsilon
#create the grid
xx = np.arange(mins[0], maxs[0])
yy = np.arange(mins[1], maxs[1])
Now you can perform a double for loop, tacking, each time, two consecutive elements of xx and yy, to do this, you can simple take:
x1 = xx[:-1] #excluding the last element
x2 = xx[1:] #excluding the first element
#the same for y:
y1 = yy[:-1] #excluding the last element
y2 = yy[1:] #excluding the first element
fluxes_grid = np.zeros((xx.shape[0], yy.shape[0]))
for i, (x1_i, x2_i) in enumerate(zip(x1, x2)):
for j, (y1_j, y2_j) in enumerate(zip(y1, y2)):
idx = np.where((x>=x1_i) & (x<x2_i) & (y>=y1_j) & (y<y2_j))[0]
fluxes_grid[i,j] = np.sum(fluxes[idx])
At the end of this loop you have a grid whose elements are pixels representing the sum of fluxes.
2 - USING A QUANTIZATION ALGORITHM LIKE K-NN
What happen if you have a lot o points, so many that the loop takes hours?
A faster solution is to use a quantization method, like K Nearest Neighbor, KNN on a rigid grid. There are many way to run a KNN (included already implemented version, e.g. sklearn KNN). But this is vary efficient if you can take advantage of a GPU. For example this my tensorflow (vs 2.1) implementation. After you have defined a squared grid:
_min, maxs = min(mins), max(maxs)
xx = np.arange(_min, _max)
yy = np.arange(_min, _max)
You can build the matrix, grid, and your position matrix, X:
grid = np.column_stack([xx, yy])
X = np.column_stack([x, y])
then you have to define a matrix euclidean pairwise-distance function:
#tf.function
def pairwise_dist(A, B):
# squared norms of each row in A and B
na = tf.reduce_sum(tf.square(A), 1)
nb = tf.reduce_sum(tf.square(B), 1)
# na as a row and nb as a co"lumn vectors
na = tf.reshape(na, [-1, 1])
nb = tf.reshape(nb, [1, -1])
# return pairwise euclidead difference matrix
D = tf.sqrt(tf.maximum(na - 2*tf.matmul(A, B, False, True) + nb, 0.0))
return D
Thus:
#compute the pairwise distances:
D = pairwise_dist(grid, X)
D = D.numpy() #get a numpy matrix from a tf tensor
#D has shape M, N, where M is the number of points in the grid and N the number of positions.
#now take a rank and from this the best K (e.g. 10)
ranks = np.argsort(D, axis=1)[:, :10]
#for each point in the grid you have the nearest ten.
Now you have to take the fluxes corresponding to this 10 positions and sum on them.
I had avoid to further specify this second method, I don't know the dimension of your catalogue, if you have or not a GPU or if you want to use such kind of optimization.
If you want I can improve this explanation, only if you are interested.

Fastest code to calculate distance between points in numpy array with cyclic (periodic) boundary conditions

I know how to calculate the Euclidean distance between points in an array using
scipy.spatial.distance.cdist
Similar to answers to this question:
Calculate Distances Between One Point in Matrix From All Other Points
However, I would like to make the calculation assuming cyclic boundary conditions, e.g. so that point [0,0] is distance 1 from point [0,n-1] in this case, not a distance of n-1. (I will then make a mask for all points within a threshold distance of my target cells, but that is not central to the question).
The only way I can think of is to repeat the calculation 9 times, with the domain indices having n added/subtracted in the x, y and then x&y directions, and then stacking the results and finding the minimum across the 9 slices. To illustrate the need for 9 repetitions, I put together a simple schematic with just 1 J-point, marked with a circle, and which shows an example where the cell marked by the triangle in this case has its nearest neighbour in the domain reflected to the top-left.
this is the code I developed for this using cdist:
import numpy as np
from scipy import spatial
n=5 # size of 2D box (n X n points)
np.random.seed(1) # to make reproducible
a=np.random.uniform(size=(n,n))
i=np.argwhere(a>-1) # all points, for each loc we want distance to nearest J
j=np.argwhere(a>0.85) # set of J locations to find distance to.
# this will be used in the KDtree soln
global maxdist
maxdist=2.0
def dist_v1(i,j):
dist=[]
# 3x3 search required for periodic boundaries.
for xoff in [-n,0,n]:
for yoff in [-n,0,n]:
jo=j.copy()
jo[:,0]-=xoff
jo[:,1]-=yoff
dist.append(np.amin(spatial.distance.cdist(i,jo,metric='euclidean'),1))
dist=np.amin(np.stack(dist),0).reshape([n,n])
return(dist)
This works, and produces e.g. :
print(dist_v1(i,j))
[[1.41421356 1. 1.41421356 1.41421356 1. ]
[2.23606798 2. 1.41421356 1. 1.41421356]
[2. 2. 1. 0. 1. ]
[1.41421356 1. 1.41421356 1. 1. ]
[1. 0. 1. 1. 0. ]]
The zeros obviously mark the J points, and the distances are correct (this EDIT corrects my earlier attempts which was incorrect).
Note that if you change the last two lines to stack the raw distances and then only use one minimum like this :
def dist_v2(i,j):
dist=[]
# 3x3 search required for periodic boundaries.
for xoff in [-n,0,n]:
for yoff in [-n,0,n]:
jo=j.copy()
jo[:,0]-=xoff
jo[:,1]-=yoff
dist.append(spatial.distance.cdist(i,jo,metric='euclidean'))
dist=np.amin(np.dstack(dist),(1,2)).reshape([n,n])
return(dist)
it is faster for small n (<10) but considerably slower for larger arrays (n>10)
...but either way, it is slow for my large arrays (N=500 and J points number around 70), this search is taking up about 99% of the calculation time, (and it is a bit ugly too using the loops) - is there a better/faster way?
The other options I thought of were:
scipy.spatial.KDTree.query_ball_point
With further searching I have found that there is a function
scipy.spatial.KDTree.query_ball_point which directly calculates the coordinates within a radius of my J-points, but it doesn't seem to have any facility to use periodic boundaries, so I presume one would still need to somehow use a 3x3 loop, stack and then use amin as I do above, so I'm not sure if this will be any faster.
I coded up a solution using this function WITHOUT worrying about the periodic boundary conditions (i.e. this doesn't answer my question)
def dist_v3(n,j):
x, y = np.mgrid[0:n, 0:n]
points = np.c_[x.ravel(), y.ravel()]
tree=spatial.KDTree(points)
mask=np.zeros([n,n])
for results in tree.query_ball_point((j), maxdist):
mask[points[results][:,0],points[results][:,1]]=1
return(mask)
Maybe I'm not using it in the most efficient way, but this is already as slow as my cdist-based solutions even without the periodic boundaries. Including the mask function in the two cdist solutions, i.e. replacing the return(dist) with return(np.where(dist<=maxdist,1,0)) in those functions, and then using timeit, I get the following timings for n=100:
from timeit import timeit
print("cdist v1:",timeit(lambda: dist_v1(i,j), number=3)*100)
print("cdist v2:",timeit(lambda: dist_v2(i,j), number=3)*100)
print("KDtree:", timeit(lambda: dist_v3(n,j), number=3)*100)
cdist v1: 181.80927299981704
cdist v2: 554.8205785999016
KDtree: 605.119637199823
Make an array of relative coordinates for points within a set distance of [0,0] and then manually loop over the J points setting up a mask with this list of relative points - This has the advantage that the "relative distance" calculation is only performed once (my J points change each timestep), but I suspect the looping will be very slow.
Precalculate a set of masks for EVERY point in the 2D domain, so in each timestep of the model integration I just pick out the mask for the J-point and apply. This would use a LOT of memory (proportional to n^4) and perhaps is still slow as you need to loop over J points to combine the masks.
I'll show an alternative approach from an image processing perspective, which may be of interest to you, regardless of whether it's the fastest or not. For convenience, I've only implemented it for an odd n.
Rather than considering a set of nxn points i, let's instead take the nxn box. We can consider this as a binary image. Let each point in j be a positive pixel in this image. For n=5 this would look something like:
Now let's think about another concept from image processing: Dilation. For any input pixel, if it has a positive pixel in its neighborhood, the output pixel will be 1. This neighborhood is defined by what is called the Structuring Element: a boolean kernel where the ones will show which neighbors to consider.
Here's how I define the SE for this problem:
Y, X = np.ogrid[-n:n+1, -n:n+1]
SQ = X*X+Y*Y
H = SQ == r
Intuitively, H is a mask denoting 'all points from the center at who satisfy the equation x*x+y*y=r. That is, all points in H lie at sqrt(r) distance from the center. Another visualization and it'll be absolutely clear:
It is an ever expanding pixel circle. Each white pixel in each mask denotes a point where the distance from the center pixel is exactly sqrt(r). You might also be able to tell that if we iteratively increase the value of r, we're actually steadily covering all the pixel locations around a particular location, eventually covering the entire image. (Some values of r don't give responses, because no such distance sqrt(r) exists for any pair of points. We skip those r values - like 3.)
So here's what the main algorithm does.
We will incrementally increase the value of r starting from 0 to some high upper limit.
At each step, if any position (x,y) in the image gives a response to dilation, that means that there is a j point at exactly sqrt(r) distance from it!
We can find a match multiple times; we'll only keep the first match and discard further matches for points. We do this till all pixels (x,y) have found their minimum distance / first match.
So you could say that this algorithm is dependent on the number of unique distance pairs in the nxn image.
This also implies that if you have more and more points in j, the algorithm will actually get faster, which goes against common sense!
The worst case for this dilation algorithm is when you have the minimum number of points (exactly one point in j), because then it would need to iterate r to a very high value to get a match from points far away.
In terms of implementing:
n=5 # size of 2D box (n X n points)
np.random.seed(1) # to make reproducible
a=np.random.uniform(size=(n,n))
I=np.argwhere(a>-1) # all points, for each loc we want distance to nearest J
J=np.argwhere(a>0.85)
Y, X = np.ogrid[-n:n+1, -n:n+1]
SQ = X*X+Y*Y
point_space = np.zeros((n, n))
point_space[J[:,0], J[:,1]] = 1
C1 = point_space[:, :n//2]
C2 = point_space[:, n//2+1:]
C = np.hstack([C2, point_space, C1])
D1 = point_space[:n//2, :]
D2 = point_space[n//2+1:, :]
D2_ = np.hstack([point_space[n//2+1:, n//2+1:],D2,point_space[n//2+1:, :n//2]])
D1_ = np.hstack([point_space[:n//2:, n//2+1:],D1,point_space[:n//2, :n//2]])
D = np.vstack([D2_, C, D1_])
p = (3*n-len(D))//2
D = np.pad(D, (p,p), constant_values=(0,0))
plt.imshow(D, cmap='gray')
plt.title(f'n={n}')
If you look at the image for n=5, you can tell what I've done; I've simply padded the image with its four quadrants in a way to represent the cyclic space, and then added some additional zero padding to account for the worst case search boundary.
#nb.jit
def dilation(image, output, kernel, N, i0, i1):
for i in range(i0,i1):
for j in range(i0, i1):
a_0 = i-(N//2)
a_1 = a_0+N
b_0 = j-(N//2)
b_1 = b_0+N
neighborhood = image[a_0:a_1, b_0:b_1]*kernel
if np.any(neighborhood):
output[i-i0,j-i0] = 1
return output
#nb.njit(cache=True)
def progressive_dilation(point_space, out, total, dist, SQ, n, N_):
for i in range(N_):
if not np.any(total):
break
H = SQ == i
rows, cols = np.nonzero(H)
if len(rows) == 0: continue
rmin, rmax = rows.min(), rows.max()
cmin, cmax = cols.min(), cols.max()
H_ = H[rmin:rmax+1, cmin:cmax+1]
out[:] = False
out = dilation(point_space, out, H_, len(H_), n, 2*n)
idx = np.logical_and(out, total)
for a, b in zip(*np.where(idx)):
dist[a, b] = i
total = total * np.logical_not(out)
return dist
def dilateWrap(D, SQ, n):
out = np.zeros((n,n), dtype=bool)
total = np.ones((n,n), dtype=bool)
dist=-1*np.ones((n,n))
dist = progressive_dilation(D, out, total, dist, SQ, n, 2*n*n+1)
return dist
dout = dilateWrap(D, SQ, n)
If we visualize dout, we can actually get an awesome visual representation of the distances.
The dark spots are basically positions where j points were present. The brightest spots naturally means points farthest away from any j. Note that I've kept the values in squared form to get an integer image. The actual distance is still one square root away. The results match with the outputs of the ball park algorithm.
# after resetting n = 501 and rerunning the first block
N = J.copy()
NE = J.copy()
E = J.copy()
SE = J.copy()
S = J.copy()
SW = J.copy()
W = J.copy()
NW = J.copy()
N[:,1] = N[:,1] - n
NE[:,0] = NE[:,0] - n
NE[:,1] = NE[:,1] - n
E[:,0] = E[:,0] - n
SE[:,0] = SE[:,0] - n
SE[:,1] = SE[:,1] + n
S[:,1] = S[:,1] + n
SW[:,0] = SW[:,0] + n
SW[:,1] = SW[:,1] + n
W[:,0] = W[:,0] + n
NW[:,0] = NW[:,0] + n
NW[:,1] = NW[:,1] - n
def distBP(I,J):
tree = BallTree(np.concatenate([J,N,E,S,W,NE,SE,SW,NW]), leaf_size=15, metric='euclidean')
dist = tree.query(I, k=1, return_distance=True)
minimum_distance = dist[0].reshape(n,n)
return minimum_distance
print(np.array_equal(distBP(I,J), np.sqrt(dilateWrap(D, SQ, n))))
Out:
True
Now for a time check, at n=501.
from timeit import timeit
nl=1
print("ball tree:",timeit(lambda: distBP(I,J),number=nl))
print("dilation:",timeit(lambda: dilateWrap(D, SQ, n),number=nl))
Out:
ball tree: 1.1706031339999754
dilation: 1.086665302000256
I would say they are roughly equal, although dilation has a very minute edge. In fact, dilation is still missing a square root operation, let's add that.
from timeit import timeit
nl=1
print("ball tree:",timeit(lambda: distBP(I,J),number=nl))
print("dilation:",timeit(lambda: np.sqrt(dilateWrap(D, SQ, n)),number=nl))
Out:
ball tree: 1.1712950239998463
dilation: 1.092416919000243
Square root basically has negligible effect on the time.
Now, I said earlier that dilation becomes faster when there are actually more points in j. So let's increase the number of points in j.
n=501 # size of 2D box (n X n points)
np.random.seed(1) # to make reproducible
a=np.random.uniform(size=(n,n))
I=np.argwhere(a>-1) # all points, for each loc we want distance to nearest J
J=np.argwhere(a>0.4) # previously a>0.85
Checking the time now:
from timeit import timeit
nl=1
print("ball tree:",timeit(lambda: distBP(I,J),number=nl))
print("dilation:",timeit(lambda: np.sqrt(dilateWrap(D, SQ, n)),number=nl))
Out:
ball tree: 3.3354218500007846
dilation: 0.2178608220001479
Ball tree has actually gotten slower while dilation got faster! This is because if there are many j points, we can quickly find all distances with a few repeats of dilation. I find this effect rather interesting - normally you would expect runtimes to get worse as number of points increase, but the opposite happens here.
Conversely, if we reduce j, we'll see dilation get slower:
#Setting a>0.9
print("ball tree:",timeit(lambda: distBP(I,J),number=nl))
print("dilation:",timeit(lambda: np.sqrt(dilateWrap(D, SQ, n)),number=nl))
Out:
ball tree: 1.010353464000218
dilation: 1.4776274510004441
I think we can safely conclude that convolutional or kernel based approaches would offer much better gains in this particular problem, rather than pairs or points or tree based approaches.
Lastly, I've mentioned it at the beginning and I'll mention it again: this entire implementation only accounts for an odd value of n; I didn't have the patience to calculate the proper padding for an even n. (If you're familiar with image processing, you've probably faced this before: everything's easier with odd sizes.)
This may also be further optimized, since I'm only an occasional dabbler in numba.
[EDIT] - I found a mistake in the way the code keeps track of the points where the job is done, fixed it with the mask_kernel. The pure python version of the newer code is ~1.5 times slower, but the numba version is slightly faster (due to some other optimisations).
[current best : ~100xto 120x the original speed]
First of all, thank you for submitting this problem, I had a lot of fun optimizing it!
My current best solution relies on the assumption that the grid is regular and that the "source" points (the ones from which we need to compute the distance) are roughly evenly distributed.
The idea here is that all of the distances are going to be either 1, sqrt(2), sqrt(3), ... so we can do the numerical calculation beforehand. Then we simply put these values in a matrix and copy that matrix around each source point (and making sure to keep the minimum value found at each point). This covers the vast majority of the points (>99%). Then we apply another more "classical" method for the remaining 1%.
Here's the code:
import numpy as np
def sq_distance(x1, y1, x2, y2, n):
# computes the pairwise squared distance between 2 sets of points (with periodicity)
# x1, y1 : coordinates of the first set of points (source)
# x2, y2 : same
dx = np.abs((np.subtract.outer(x1, x2) + n//2)%(n) - n//2)
dy = np.abs((np.subtract.outer(y1, y2) + n//2)%(n) - n//2)
d = (dx*dx + dy*dy)
return d
def apply_kernel(sources, sqdist, kern_size, n, mask):
ker_i, ker_j = np.meshgrid(np.arange(-kern_size, kern_size+1), np.arange(-kern_size, kern_size+1), indexing="ij")
kernel = np.add.outer(np.arange(-kern_size, kern_size+1)**2, np.arange(-kern_size, kern_size+1)**2)
mask_kernel = kernel > kern_size**2
for pi, pj in sources:
ind_i = (pi+ker_i)%n
ind_j = (pj+ker_j)%n
sqdist[ind_i,ind_j] = np.minimum(kernel, sqdist[ind_i,ind_j])
mask[ind_i,ind_j] *= mask_kernel
def dist_vf(sources, n, kernel_size):
sources = np.asfortranarray(sources) #for memory contiguity
kernel_size = min(kernel_size, n//2)
kernel_size = max(kernel_size, 1)
sqdist = np.full((n,n), 10*n**2, dtype=np.int32) #preallocate with a huge distance (>max**2)
mask = np.ones((n,n), dtype=bool) #which points have not been reached?
#main code
apply_kernel(sources, sqdist, kernel_size, n, mask)
#remaining points
rem_i, rem_j = np.nonzero(mask)
if len(rem_i) > 0:
sq_d = sq_distance(sources[:,0], sources[:,1], rem_i, rem_j, n).min(axis=0)
sqdist[rem_i, rem_j] = sq_d
#eff = 1-rem_i.size/n**2
#print("covered by kernel :", 100*eff, "%")
#print("overlap :", sources.shape[0]*(1+2*kernel_size)**2/n**2)
#print()
return np.sqrt(sqdist)
Testing this version with
n=500 # size of 2D box (n X n points)
np.random.seed(1) # to make reproducible
a=np.random.uniform(size=(n,n))
all_points=np.argwhere(a>-1) # all points, for each loc we want distance to nearest J
source_points=np.argwhere(a>1-70/n**2) # set of J locations to find distance to.
#
# code for dist_v1 and dist_vf
#
overlap=5.2
kernel_size = int(np.sqrt(overlap*n**2/source_points.shape[0])/2)
print("cdist v1 :", timeit(lambda: dist_v1(all_points,source_points), number=1)*1000, "ms")
print("kernel version:", timeit(lambda: dist_vf(source_points, n, kernel_size), number=10)*100, "ms")
gives
cdist v1 : 1148.6694 ms
kernel version: 69.21876999999998 ms
which is a already a ~17x speedup! I also implemented a numba version of sq_distance and apply_kernel: [this is the new correct version]
#njit(cache=True)
def sq_distance(x1, y1, x2, y2, n):
m1 = x1.size
m2 = x2.size
n2 = n//2
d = np.empty((m1,m2), dtype=np.int32)
for i in range(m1):
for j in range(m2):
dx = np.abs(x1[i] - x2[j] + n2)%n - n2
dy = np.abs(y1[i] - y2[j] + n2)%n - n2
d[i,j] = (dx*dx + dy*dy)
return d
#njit(cache=True)
def apply_kernel(sources, sqdist, kern_size, n, mask):
# creating the kernel
kernel = np.empty((2*kern_size+1, 2*kern_size+1))
vals = np.arange(-kern_size, kern_size+1)**2
for i in range(2*kern_size+1):
for j in range(2*kern_size+1):
kernel[i,j] = vals[i] + vals[j]
mask_kernel = kernel > kern_size**2
I = sources[:,0]
J = sources[:,1]
# applying the kernel for each point
for l in range(sources.shape[0]):
pi = I[l]
pj = J[l]
if pj - kern_size >= 0 and pj + kern_size<n: #if we are in the middle, no need to do the modulo for j
for i in range(2*kern_size+1):
ind_i = np.mod((pi+i-kern_size), n)
for j in range(2*kern_size+1):
ind_j = (pj+j-kern_size)
sqdist[ind_i,ind_j] = np.minimum(kernel[i,j], sqdist[ind_i,ind_j])
mask[ind_i,ind_j] = mask_kernel[i,j] and mask[ind_i,ind_j]
else:
for i in range(2*kern_size+1):
ind_i = np.mod((pi+i-kern_size), n)
for j in range(2*kern_size+1):
ind_j = np.mod((pj+j-kern_size), n)
sqdist[ind_i,ind_j] = np.minimum(kernel[i,j], sqdist[ind_i,ind_j])
mask[ind_i,ind_j] = mask_kernel[i,j] and mask[ind_i,ind_j]
return
and testing with
overlap=5.2
kernel_size = int(np.sqrt(overlap*n**2/source_points.shape[0])/2)
print("cdist v1 :", timeit(lambda: dist_v1(all_points,source_points), number=1)*1000, "ms")
print("kernel numba (first run):", timeit(lambda: dist_vf(source_points, n, kernel_size), number=1)*1000, "ms") #first run = cimpilation = long
print("kernel numba :", timeit(lambda: dist_vf(source_points, n, kernel_size), number=10)*100, "ms")
which gave the following results
cdist v1 : 1163.0742 ms
kernel numba (first run): 2060.0802 ms
kernel numba : 8.80377000000001 ms
Due to the JIT compilation, the first run is pretty slow but otherwise, it's a 120x improvement!
It may be possible to get a little bit more out of this algorithm by tweaking the kernel_size parameter (or the overlap). The current choice of kernel_size is only effective for a small number of source points. For example, this choice fails miserably with source_points=np.argwhere(a>0.85) (13s) while manually setting kernel_size=5 gives the answer in 22ms.
I hope my post isn't (unnecessarily) too complicated, I don't really know how to organise it better.
[EDIT 2]:
I gave a little more attention to the non-numba part of the code and managed to get a pretty significant speedup, getting very close to what numba could achieve: Here is the new version of the function apply_kernel:
def apply_kernel(sources, sqdist, kern_size, n, mask):
ker_i = np.arange(-kern_size, kern_size+1).reshape((2*kern_size+1,1))
ker_j = np.arange(-kern_size, kern_size+1).reshape((1,2*kern_size+1))
kernel = np.add.outer(np.arange(-kern_size, kern_size+1)**2, np.arange(-kern_size, kern_size+1)**2)
mask_kernel = kernel > kern_size**2
for pi, pj in sources:
imin = pi-kern_size
jmin = pj-kern_size
imax = pi+kern_size+1
jmax = pj+kern_size+1
if imax < n and jmax < n and imin >=0 and jmin >=0: # we are inside
sqdist[imin:imax,jmin:jmax] = np.minimum(kernel, sqdist[imin:imax,jmin:jmax])
mask[imin:imax,jmin:jmax] *= mask_kernel
elif imax < n and imin >=0:
ind_j = (pj+ker_j.ravel())%n
sqdist[imin:imax,ind_j] = np.minimum(kernel, sqdist[imin:imax,ind_j])
mask[imin:imax,ind_j] *= mask_kernel
elif jmax < n and jmin >=0:
ind_i = (pi+ker_i.ravel())%n
sqdist[ind_i,jmin:jmax] = np.minimum(kernel, sqdist[ind_i,jmin:jmax])
mask[ind_i,jmin:jmax] *= mask_kernel
else :
ind_i = (pi+ker_i)%n
ind_j = (pj+ker_j)%n
sqdist[ind_i,ind_j] = np.minimum(kernel, sqdist[ind_i,ind_j])
mask[ind_i,ind_j] *= mask_kernel
The main optimisations are
Indexing with slices (rather than a dense array)
Use of sparse indexes (how did I not think about that earlier)
Testing with
overlap=5.4
kernel_size = int(np.sqrt(overlap*n**2/source_points.shape[0])/2)
print("cdist v1 :", timeit(lambda: dist_v1(all_points,source_points), number=1)*1000, "ms")
print("kernel v2 :", timeit(lambda: dist_vf(source_points, n, kernel_size), number=10)*100, "ms")
gives
cdist v1 : 1209.8163000000002 ms
kernel v2 : 11.319049999999997 ms
which is a nice 100x improvement over cdist, a ~5.5x improvement over the previous numpy-only version and just ~25% slower than what I could achieve with numba.
Here are a fixed version of your code and a different method that is a bit faster. They give the same results so I'm reasonably confident they are correct:
import numpy as np
from scipy.spatial.distance import squareform, pdist, cdist
from numpy.linalg import norm
def pb_OP(A, p=1.0):
distl = []
for *offs, ct in [(0, 0, 0), (0, p, 1), (p, 0, 1), (p, p, 1), (-p, p, 1)]:
B = A - offs
distl.append(cdist(B, A, metric='euclidean'))
if ct:
distl.append(distl[-1].T)
return np.amin(np.dstack(distl), 2)
def pb_pp(A, p=1.0):
out = np.empty((2, A.shape[0]*(A.shape[0]-1)//2))
for o, i in zip(out, A.T):
pdist(i[:, None], 'cityblock', out=o)
out[out > p/2] -= p
return squareform(norm(out, axis=0))
test = np.random.random((1000, 2))
assert np.allclose(pb_OP(test), pb_pp(test))
from timeit import timeit
t_OP = timeit(lambda: pb_OP(test), number=10)*100
t_pp = timeit(lambda: pb_pp(test), number=10)*100
print('OP', t_OP)
print('pp', t_pp)
Sample run. 1000 points:
OP 210.11001259903423
pp 22.288734700123314
We see that my method is ~9x faster which by a neat coincidence is the number of offset cponfigurations OP's version has to check. It uses pdist on the individual coordinates to get absolute differences. Where these are larger than half the grid spacing we subtract one period. It remains to take Euclidean norm and to unpack storage.
For calculating multiple distances I think it is hard to beat a simple BallTree (or similar).
I didn't quite understand the cyclic boundary, or at least why you need to loop 3x3 times, as I see it is behaves like torus and it is enough to make 5 copies.
Update: Indeed you need 3x3 for the edges. I updated the code.
To make sure my minimum_distance is correct by doing for n = 200 a np.all( minimum_distance == dist_v1(i,j) ) test gave True.
For n = 500 generated with the provided code, the %%time for a cold start gave
CPU times: user 1.12 s, sys: 0 ns, total: 1.12 s
Wall time: 1.11 s
So I generate 500 data points like in the post
import numpy as np
n=500 # size of 2D box (n X n points)
np.random.seed(1) # to make reproducible
a=np.random.uniform(size=(n,n))
i=np.argwhere(a>-1) # all points, for each loc we want distance to nearest J
j=np.argwhere(a>0.85) # set of J locations to find distance to.
And use the BallTree
import numpy as np
from sklearn.neighbors import BallTree
N = j.copy()
NE = j.copy()
E = j.copy()
SE = j.copy()
S = j.copy()
SW = j.copy()
W = j.copy()
NW = j.copy()
N[:,1] = N[:,1] - n
NE[:,0] = NE[:,0] - n
NE[:,1] = NE[:,1] - n
E[:,0] = E[:,0] - n
SE[:,0] = SE[:,0] - n
SE[:,1] = SE[:,1] + n
S[:,1] = S[:,1] + n
SW[:,0] = SW[:,0] + n
SW[:,1] = SW[:,1] + n
W[:,0] = W[:,0] + n
NW[:,0] = NW[:,0] + n
NW[:,1] = NW[:,1] - n
tree = BallTree(np.concatenate([j,N,E,S,W,NE,SE,SW,NW]), leaf_size=15, metric='euclidean')
dist = tree.query(i, k=1, return_distance=True)
minimum_distance = dist[0].reshape(n,n)
Update:
Note here I copied the data to N,E,S,W,NE,SE,NW,SE, of the box to handle the boundary conditions. Again, for n = 200 this gave the same results. You could tweak the leaf_size, but I feel this setting is allright.
The performance is sensitive for the number of points in j.
These are 8 different solutions I've timed, some of my own and some posted in response to my question, that use 4 broad approaches:
spatial cdist
spatial KDtree
Sklearn BallTree
Kernel approach
This is the code with the 8 test routines:
import numpy as np
from scipy import spatial
from sklearn.neighbors import BallTree
n=500 # size of 2D box
f=200./(n*n) # first number is rough number of target cells...
np.random.seed(1) # to make reproducable
a=np.random.uniform(size=(n,n))
i=np.argwhere(a>-1) # all points, we want to know distance to nearest point
j=np.argwhere(a>1.0-f) # set of locations to find distance to.
# long array of 3x3 j points:
for xoff in [0,n,-n]:
for yoff in [0,-n,n]:
if xoff==0 and yoff==0:
j9=j.copy()
else:
jo=j.copy()
jo[:,0]+=xoff
jo[:,1]+=yoff
j9=np.vstack((j9,jo))
global maxdist
maxdist=10
overlap=5.2
kernel_size=int(np.sqrt(overlap*n**2/j.shape[0])/2)
print("no points",len(j))
# repear cdist over each member of 3x3 block
def dist_v1(i,j):
dist=[]
# 3x3 search required for periodic boundaries.
for xoff in [-n,0,n]:
for yoff in [-n,0,n]:
jo=j.copy()
jo[:,0]+=xoff
jo[:,1]+=yoff
dist.append(np.amin(spatial.distance.cdist(i,jo,metric='euclidean'),1))
dist=np.amin(np.stack(dist),0).reshape([n,n])
#dmask=np.where(dist<=maxdist,1,0)
return(dist)
# same as v1, but taking one amin function at the end
def dist_v2(i,j):
dist=[]
# 3x3 search required for periodic boundaries.
for xoff in [-n,0,n]:
for yoff in [-n,0,n]:
jo=j.copy()
jo[:,0]+=xoff
jo[:,1]+=yoff
dist.append(spatial.distance.cdist(i,jo,metric='euclidean'))
dist=np.amin(np.dstack(dist),(1,2)).reshape([n,n])
#dmask=np.where(dist<=maxdist,1,0)
return(dist)
# using a KDTree query ball points, looping over j9 points as in online example
def dist_v3(n,j):
x,y=np.mgrid[0:n,0:n]
points=np.c_[x.ravel(), y.ravel()]
tree=spatial.KDTree(points)
mask=np.zeros([n,n])
for results in tree.query_ball_point((j), 2.1):
mask[points[results][:,0],points[results][:,1]]=1
return(mask)
# using ckdtree query on the j9 long array
def dist_v4(i,j):
tree=spatial.cKDTree(j)
dist,minid=tree.query(i)
return(dist.reshape([n,n]))
# back to using Cdist, but on the long j9 3x3 array, rather than on each element separately
def dist_v5(i,j):
# 3x3 search required for periodic boundaries.
dist=np.amin(spatial.distance.cdist(i,j,metric='euclidean'),1)
#dmask=np.where(dist<=maxdist,1,0)
return(dist)
def dist_v6(i,j):
tree = BallTree(j,leaf_size=5,metric='euclidean')
dist = tree.query(i, k=1, return_distance=True)
mindist = dist[0].reshape(n,n)
return(mindist)
def sq_distance(x1, y1, x2, y2, n):
# computes the pairwise squared distance between 2 sets of points (with periodicity)
# x1, y1 : coordinates of the first set of points (source)
# x2, y2 : same
dx = np.abs((np.subtract.outer(x1, x2) + n//2)%(n) - n//2)
dy = np.abs((np.subtract.outer(y1, y2) + n//2)%(n) - n//2)
d = (dx*dx + dy*dy)
return d
def apply_kernel1(sources, sqdist, kern_size, n, mask):
ker_i, ker_j = np.meshgrid(np.arange(-kern_size, kern_size+1), np.arange(-kern_size, kern_size+1), indexing="ij")
kernel = np.add.outer(np.arange(-kern_size, kern_size+1)**2, np.arange(-kern_size, kern_size+1)**2)
mask_kernel = kernel > kern_size**2
for pi, pj in sources:
ind_i = (pi+ker_i)%n
ind_j = (pj+ker_j)%n
sqdist[ind_i,ind_j] = np.minimum(kernel, sqdist[ind_i,ind_j])
mask[ind_i,ind_j] *= mask_kernel
def apply_kernel2(sources, sqdist, kern_size, n, mask):
ker_i = np.arange(-kern_size, kern_size+1).reshape((2*kern_size+1,1))
ker_j = np.arange(-kern_size, kern_size+1).reshape((1,2*kern_size+1))
kernel = np.add.outer(np.arange(-kern_size, kern_size+1)**2, np.arange(-kern_size, kern_size+1)**2)
mask_kernel = kernel > kern_size**2
for pi, pj in sources:
imin = pi-kern_size
jmin = pj-kern_size
imax = pi+kern_size+1
jmax = pj+kern_size+1
if imax < n and jmax < n and imin >=0 and jmin >=0: # we are inside
sqdist[imin:imax,jmin:jmax] = np.minimum(kernel, sqdist[imin:imax,jmin:jmax])
mask[imin:imax,jmin:jmax] *= mask_kernel
elif imax < n and imin >=0:
ind_j = (pj+ker_j.ravel())%n
sqdist[imin:imax,ind_j] = np.minimum(kernel, sqdist[imin:imax,ind_j])
mask[imin:imax,ind_j] *= mask_kernel
elif jmax < n and jmin >=0:
ind_i = (pi+ker_i.ravel())%n
sqdist[ind_i,jmin:jmax] = np.minimum(kernel, sqdist[ind_i,jmin:jmax])
mask[ind_i,jmin:jmax] *= mask_kernel
else :
ind_i = (pi+ker_i)%n
ind_j = (pj+ker_j)%n
sqdist[ind_i,ind_j] = np.minimum(kernel, sqdist[ind_i,ind_j])
mask[ind_i,ind_j] *= mask_kernel
def dist_v7(sources, n, kernel_size,method):
sources = np.asfortranarray(sources) #for memory contiguity
kernel_size = min(kernel_size, n//2)
kernel_size = max(kernel_size, 1)
sqdist = np.full((n,n), 10*n**2, dtype=np.int32) #preallocate with a huge distance (>max**2)
mask = np.ones((n,n), dtype=bool) #which points have not been reached?
#main code
if (method==1):
apply_kernel1(sources, sqdist, kernel_size, n, mask)
else:
apply_kernel2(sources, sqdist, kernel_size, n, mask)
#remaining points
rem_i, rem_j = np.nonzero(mask)
if len(rem_i) > 0:
sq_d = sq_distance(sources[:,0], sources[:,1], rem_i, rem_j, n).min(axis=0)
sqdist[rem_i, rem_j] = sq_d
return np.sqrt(sqdist)
from timeit import timeit
nl=10
print ("-----------------------")
print ("Timings for ",nl,"loops")
print ("-----------------------")
print("1. cdist looped amin:",timeit(lambda: dist_v1(i,j),number=nl))
print("2. cdist single amin:",timeit(lambda: dist_v2(i,j),number=nl))
print("3. KDtree ball pt:", timeit(lambda: dist_v3(n,j9),number=nl))
print("4. KDtree query:",timeit(lambda: dist_v4(i,j9),number=nl))
print("5. cdist long list:",timeit(lambda: dist_v5(i,j9),number=nl))
print("6. ball tree:",timeit(lambda: dist_v6(i,j9),number=nl))
print("7. kernel orig:", timeit(lambda: dist_v7(j, n, kernel_size,1), number=nl))
print("8. kernel optimised:", timeit(lambda: dist_v7(j, n, kernel_size,2), number=nl))
The output (timing in seconds) on my linux 12 core desktop (with 48GB RAM) for n=350 and 63 points:
no points 63
-----------------------
Timings for 10 loops
-----------------------
1. cdist looped amin: 3.2488364999881014
2. cdist single amin: 6.494611179979984
3. KDtree ball pt: 5.180531410995172
4. KDtree query: 0.9377906009904109
5. cdist long list: 3.906166430999292
6. ball tree: 3.3540162370190956
7. kernel orig: 0.7813036740117241
8. kernel optimised: 0.17046571199898608
and for n=500 and npts=176:
no points 176
-----------------------
Timings for 10 loops
-----------------------
1. cdist looped amin: 16.787221198988846
2. cdist single amin: 40.97849371898337
3. KDtree ball pt: 9.926229109987617
4. KDtree query: 0.8417396580043714
5. cdist long list: 14.345821461000014
6. ball tree: 1.8792325239919592
7. kernel orig: 1.0807358759921044
8. kernel optimised: 0.5650744160229806
So in summary I reached the following conclusions:
avoid cdist if you have quite a large problem
If your problem is not too computational-time constrained I would recommend the "KDtree query" approach as it is just 2 lines without the periodic boundaries and a few more with periodic boundary to set up the j9 array
For maximum performance (e.g. a long integration of a model where this is required each time step as is my case) then the Kernal solution is now by far the fastest.

Apply DFT matrix along each axis of 3D array in NumPy?

I can first obtain the DFT matrix of a given size, say n by
import numpy as np
n = 64
D = np.fft.fft(np.eye(n))
The FFT is of course just a quick algorithm for applying D to a vector:
x = np.random.randn(n)
ft1 = np.dot(D,x)
print( np.abs(ft1 - fft.fft(x)).max() )
# prints near double precision roundoff
The 2D FFT can be obtained by applying D to both the rows and columns of a matrix:
x = np.random.randn(n,n)
ft2 = np.dot(x, D.T) # Apply D to rows.
ft2 = np.dot(D, ft2) # Apply D to cols.
print( np.abs(ft2 - fft.fft2(x)).max() )
# near machine round off again
How do I compute this analogously for the 3 dimensional Discrete Fourier Transform?
I.e.,
x = np.random.randn(n,n,n)
ft3 = # dot operations using D and x
print( np.abs(ft3 - fft.fftn(x)).max() )
# prints near zero
Essentially, I think I need to apply D to each column vector in the volume, then each row vector in the volume, and finally each "depth vector". But I'm not sure how to do this using dot.
You can use the einsum expression to perform the transformation on each index:
x = np.random.randn(n, n, n)
ft3 = np.einsum('ijk,im->mjk', x, D)
ft3 = np.einsum('ijk,jm->imk', ft3, D)
ft3 = np.einsum('ijk,km->ijm', ft3, D)
print(np.abs(ft3 - np.fft.fftn(x)).max())
1.25571216554e-12
This can also be written as a single NumPy step:
ft3 = np.einsum('ijk,im,jn,kl->mnl', ft3, D, D, D, optimize=True)
Without the optimize argument (available in NumPy 1.12+) it will be very slow however. You can also do each of the steps using dot, but it requires a bit of reshaping and transposing. In NumPy 1.14+ the einsum function will automatically detect the BLAS operations and do this for you.

Adding Euclidean distance to a matrix

I explain what I have to develop.
Let's say I have to perform a function that is responsible for receiving two matrices, which have the same number of columns but can differ in the number of rows.
In summary, we will have two matrices of vectors with the same dimension but different number N of elements.
I have to calculate the Euclidean distance between each of the vectors that make up my two matrices, and then store it in another matrix that will contain the Euclidean distance between all my vectors.
This is the code I have developed:
def compute_distances(x, y):
# Dimension:
N, d = x.shape
M, d_ = y.shape
# The dimension should be the same
if d != d_:
print "Dimensiones de x e y no coinciden, no puedo calcular las distancias..."
return None
# Calculate distance with loops:
D = np.zeros((N, M))
i = 0
j = 0
for v1 in x:
for v2 in y:
if(j != M):
D[i,j] = math.sqrt(sum([(xi-yi)**2 for xi,yi in zip(v1,v2)]))
#print "[",i,",",j,"]"
j = j + 1
else:
j = 0
i = i + 1;
print D
In this method I am receiving the two matrices to later create a matrix that will have the Euclidean distances between the vectors of my matrices x and y.
The problem is the following, I do not know how, to each one of the calculated Euclidean distance values ​​I have to assign the correct position of the new matrix D that I have generated.
My main function has the following structure:
n = 1000
m = 700
d = 10
x = np.random.randn(n, d)
y = np.random.randn(m, d)
print "x shape =", x.shape
print "y shape =", y.shape
D_bucle = da.compute_distances(x, y)
D_cdist = cdist(x, y)
print np.max(np.abs(D_cdist - D_bucle))
B_cdist calculates the Euclidean distance using efficient methods.
It has to have the same result as D_bucle that calculates the same as the other but with non efficient code, but I'm not getting what the result should be.
I think it's when I create my Euclidean matrix D that is not doing it correctly, then the calculations are incorrect.
Updated!!!
I just updated my solution, my problem is that firstly I didnt know how to asign to the D Matrix my correct euclidean vector result for each pair of vectors,
Now I khow how to asign it but now my problem is that only the first line from D Matrix is having a correct result in comparison with cdist function
not fully understanding what you're asking, but I do see one problem which may explain your results:
for v1 in x:
for v2 in y:
D = math.sqrt(sum([(xi-yi)**2 for xi,yi in zip(v1,v2)]))
You are overwriting the value of D each of the NxM times you go through this loop. When you're done D only contains the distance of the last compare. You might need something like D[i,j] = math.sqrt(...

Distance calculation on matrix using numpy

I am trying to implement a K-means algorithm in Python (I know there is libraries for that, but I want to learn how to implement it myself.) Here is the function I am havin problem with:
def AssignPoints(points, centroids):
"""
Takes two arguments:
points is a numpy array such that points.shape = m , n where m is number of examples,
and n is number of dimensions.
centroids is numpy array such that centroids.shape = k , n where k is number of centroids.
k < m should hold.
Returns:
numpy array A such that A.shape = (m,) and A[i] is index of the centroid which points[i] is assigned to.
"""
m ,n = points.shape
temp = []
for i in xrange(n):
temp.append(np.subtract.outer(points[:,i],centroids[:,i]))
distances = np.hypot(*temp)
return distances.argmin(axis=1)
Purpose of this function, given m points in n dimensional space, and k centroids in n dimensional space, produce a numpy array of (x1 x2 x3 x4 ... xm) where x1 is the index of centroid which is closest to first point. This was working fine, until I tried it with 4 dimensional examples. When I try to put 4 dimensional examples, I get this error:
File "/path/to/the/kmeans.py", line 28, in AssignPoints
distances = np.hypot(*temp)
ValueError: invalid number of arguments
How can I fix this, or if I can't, how do you suggest I calculate what I am trying to calculate here?
My Answer
def AssignPoints(points, centroids):
m ,n = points.shape
temp = []
for i in xrange(n):
temp.append(np.subtract.outer(points[:,i],centroids[:,i]))
for i in xrange(len(temp)):
temp[i] = temp[i] ** 2
distances = np.add.reduce(temp) ** 0.5
return distances.argmin(axis=1)
Try this:
np.sqrt(((points[np.newaxis] - centroids[:,np.newaxis]) ** 2).sum(axis=2)).argmin(axis=0)
Or:
diff = points[np.newaxis] - centroids[:,np.newaxis]
norm = np.sqrt((diff*diff).sum(axis=2))
closest = norm.argmin(axis=0)
And don't ask what's it doing :D
Edit: nah, just kidding. The broadcasting in the middle (points[np.newaxis] - centroids[:,np.newaxis]) is "making" two 3D arrays from the original ones. The result is such that each "plane" contains the difference between all the points and one of the centroids. Let's call it diffs.
Then we do the usual operation to calculate the euclidean distance (square root of the squares of differences): np.sqrt((diffs ** 2).sum(axis=2)). We end up with a (k, m) matrix where row 0 contain the distances to centroids[0], etc. So, the .argmin(axis=0) gives you the result you wanted.
You need to define a distance function where you are using hypot. Usually in K-means it is
Distance=sum((point-centroid)^2)
Here is some matlab code that does it ... I can port it if you can't, but give it a go. Like you said, only way to learn.
function idx = findClosestCentroids(X, centroids)
%FINDCLOSESTCENTROIDS computes the centroid memberships for every example
% idx = FINDCLOSESTCENTROIDS (X, centroids) returns the closest centroids
% in idx for a dataset X where each row is a single example. idx = m x 1
% vector of centroid assignments (i.e. each entry in range [1..K])
%
% Set K
K = size(centroids, 1);
[numberOfExamples numberOfDimensions] = size(X);
% You need to return the following variables correctly.
idx = zeros(size(X,1), 1);
% Go over every example, find its closest centroid, and store
% the index inside idx at the appropriate location.
% Concretely, idx(i) should contain the index of the centroid
% closest to example i. Hence, it should be a value in the
% range 1..K
%
for loop=1:numberOfExamples
Distance = sum(bsxfun(#minus,X(loop,:),centroids).^2,2);
[value index] = min(Distance);
idx(loop) = index;
end;
end
UPDATE
This should return the distance, notice that the above matlab code just returns the distance(and index) of the closest centroid...your function returns all distances, as does the one below.
def FindDistance(X,centroids):
K=shape(centroids)[0]
examples, dimensions = shape(X)
distance = zeros((examples,K))
for ex in xrange(examples):
distance[ex,:] = np.sum((X[ex,:]-centroids)**2,1)
return distance

Categories

Resources