I'm attempting to convert a double summation formula into code, but can't figure out the correct matrix/vector representation of it.
The first summation is i to n, and the second is over j > i to n.
I'm guessing there is a much more efficient & pythonic way of writing this?
I resorted to nested for loops to just get it working but, as expected, it runs very slowly with a large dataset:
def wapc_denom(weights, vols):
x = []
y = []
for i, wi in enumerate(weights):
for j, wj in enumerate(weights):
if j > i:
x.append(wi * wj * vols[i] * vols[j])
y.append(np.sum(x))
return np.sum(y)
Edit:
Using guidance from smci's answer I think I have a potential solution:
def wapc_denom2(weights, vols):
return np.sum(np.tril(np.outer(weights, vols.T)**2, k=-1))
Assuming you want to count every term only once (for that you have to move the x = [] into the outer loop) one cheap way of computing the sum would be
Create mock data
weights = np.random.random(10)
vols = np.random.random(10)
Do the calculation
wv = weights * vols
result = (wv.sum()**2 - wv#wv) / 2
Check that it's the same
def wapc_denom(weights, vols):
y = []
for i, wi in enumerate(weights):
x = []
for j, wj in enumerate(weights):
if j > i:
x.append(wi * wj * vols[i] * vols[j])
y.append(np.sum(x))
return np.sum(y)
assert np.allclose(result, wapc_denom(weights, vols))
Why does it work?
What we are doing is compute the sum of the full matrix, subtract the diagonal and divide by two. This is cheap because it is easy to verify that the sum of an outer product is just the product of the summed factors.
wi * wj * vols[i] * vols[j] is a telltale. vols is another vector, so first you want to compute the vector wv = w * vols
then (wj * vols[j]) * (wi * vols[i]) = wv^T * wv is your (matrix outer product) expression; that's a column vector * a row vector. But actually you only want the sum. So I don't see a need to construct a vector y.append(np.sum(x)), you're only going to sum it anyway np.sum(y)
also the if j > i part means you only want the sum of the Lower Triangular part, and exclude the diagonal.
EDIT: the result is fully determined just from wv, I didn't think we needed the matrix to get the sum, and we didn't need the diagonal; #PaulPanzer found the most compact expression.
You can use triangulations in numpy, check np.triu and np.meshgrid. Do:
np.product(np.triu(np.meshgrid(weights,weights), 1) * np.triu(np.meshgrid(vols,vols), 1),0).sum(1).cumsum().sum()
Example:
w = np.arange(4) +1
v = np.array([1,3,2,2])
print(np.triu(np.meshgrid(w,w), k=1))
>>array([[[0, 2, 3, 4],
[0, 0, 3, 4],
[0, 0, 0, 4],
[0, 0, 0, 0]],
[[0, 1, 1, 1],
[0, 0, 2, 2],
[0, 0, 0, 3],
[0, 0, 0, 0]]])
# example of product + triu + meshgrid (your x values):
print(np.product(np.triu(np.meshgrid(w,w), 1) * np.triu(np.meshgrid(v,v), 1),0))
>>array([[ 0, 6, 6, 8],
[ 0, 0, 36, 48],
[ 0, 0, 0, 48],
[ 0, 0, 0, 0]])
print(np.product(np.triu(np.meshgrid(w,w), 1) * np.triu(np.meshgrid(v,v), 1),0).sum(1).cumsum().sum())
>> 428
print(wapc_denom(w, v))
>> 428
Related
I am looking for a way to speed up the specific operation on tensors in PyTorch. Since it is a general operation on matrices, I am open to answers in NumPy as well.
Let's say I have a tensor with values from 0 to N-1 (N=4) where each value repeats the same number of times (R=2).
import torch
x = torch.Tensor([0, 0, 1, 1, 2, 2, 3, 3])
In this case, it is sorted, but any permutation of x is also in the set of considered tensors X.
I am getting an input tensor with values from 0 to N-1 but without any constraints on the repetition.
z = torch.tensor([3, 2, 3, 0, 2, 3, 1, 2])
And I would like to find an efficient implementation of foo such that y = foo(z). y should be some permutation of x (from the set X) that tries to do as few changes in z as possible (in terms of Hamming distance), for example
y = torch.tensor([3, 2, 3, 0, 2, 0, 1, 1])
The trivial solution is to keep counting the number elements with the same value, but it is extremely inefficient to process elements one-by-one for larger tensors:
def foo(z):
R = 2
N = 4
counters = [0] * N
# first, we replace extra elements with -1
y = []
for elem in z:
if counters[elem] < R:
counters[elem] += 1
y.append(elem)
else:
y.append(-1)
y = torch.tensor(y)
assert torch.equal(y, torch.tensor([3, 2, 3, 0, 2, -1, 1, -1]))
# second, we replace -1 by "unfilled" counters
for i in range(len(y)):
if y[i] == -1:
first_unfilled = [n for n in range(N) if counters[n] < R][0]
counters[first_unfilled] += 1
y[i] = first_unfilled
return y
assert torch.equal(y, foo(z))
When I multiply two big integers using FFT, I find the result of FFT and IFFT is always not right.
method
To realize FFT, I just follow the pseudocode as followed:
the pseudocode of FFT
The equations of FFT and IFFT are as followed. So, when realizing IFFT, I just replace a with y, replace omega with omega ^^ -1 and divide it by n. And, use flag to distinguish them in my function.
For FFT, y will be
For IFFT, a will be
problem
To find the problem, I try to compare the results between numpy.fft and my function.
FFT.
The results of numpy and my function look the same, but the sign of images is the opposite. For example (the second element of case2 below):
my function result: -4-9.65685424949238j
numpy result: -4+9.65685424949238j
IFFT. I just find it wrong, and can't find any rule.
python code
Here is my function FFT, and comparison:
from typing import List
from cmath import pi, exp
from numpy.fft import fft, ifft
def FFT(a: List, flag: bool) -> List:
"""realize DFT using FFT"""
n = len(a)
if n == 1:
return a
# complex root
omg_n = exp(2 * pi * 1j / n)
if flag:
# IFFT
omg_n = 1 / omg_n
omg = 1
# split a into 2 part
a0 = a[::2] # even
a1 = a[1::2] # odd
# corresponding y
y0 = FFT(a0, flag)
y1 = FFT(a1, flag)
# result y
y = [0] * n
for k in range(n // 2):
y[k] = y0[k] + omg * y1[k]
y[k + n // 2] = y0[k] - omg * y1[k]
omg = omg * omg_n
# IFFT
if flag:
y = [i / n for i in y]
return y
if __name__ == '__main__':
test_cases = [
[1, 1],
[1, 2, 3, 4, 5, 6, 7, 8],
[1, 4, 2, 9, 0, 0, 3, 8, 9, 1, 4, 0, 0, 0, 0, 0, ],
]
print("test FFT")
for i, case in enumerate(test_cases):
print(f"case{i + 1}", case)
manual_result = FFT(case, False)
numpy_result = fft(case).tolist()
print("manual_result:", manual_result)
print("numpy_result:", numpy_result)
print("difference:", [i - j for i, j in zip(manual_result, numpy_result)])
print()
print("test IFFT")
for i, case in enumerate(test_cases):
print(f"case{i + 1}", case)
manual_result = FFT(case, True)
numpy_result = ifft(case).tolist()
print("manual_result:", manual_result)
print("numpy_result:", numpy_result)
print("difference:", [i - j for i, j in zip(manual_result, numpy_result)])
print()
The FFT output:
test FFT
case1 [1, 1]
manual_result: [2, 0]
numpy_result: [(2+0j), 0j]
difference: [0j, 0j]
case2 [1, 2, 3, 4, 5, 6, 7, 8]
manual_result: [36, (-4-9.65685424949238j), (-4-4.000000000000001j), (-4-1.6568542494923815j), -4, (-4+1.6568542494923806j), (-4+4.000000000000001j), (-3.999999999999999+9.656854249492381j)]
numpy_result: [(36+0j), (-4+9.65685424949238j), (-4+4j), (-4+1.6568542494923806j), (-4+0j), (-4-1.6568542494923806j), (-4-4j), (-4-9.65685424949238j)]
difference: [0j, -19.31370849898476j, -8j, -3.313708498984762j, 0j, 3.313708498984761j, 8j, (8.881784197001252e-16+19.31370849898476j)]
case3 [1, 4, 2, 9, 0, 0, 3, 8, 9, 1, 4, 0, 0, 0, 0, 0]
manual_result: [41, (-12.710780677203363+13.231540329804117j), (12.82842712474619+7.2426406871192865j), (-14.692799048494296+7.4256307475248935j), (1.0000000000000013-12j), (5.763866860359768+6.0114171851517995j), (7.171572875253808+1.2426406871192839j), (-10.360287134662114+11.817326767431025j), -3, (-10.360287134662112-11.817326767431021j), (7.17157287525381-1.2426406871192848j), (5.763866860359771-6.011417185151798j), (0.9999999999999987+12j), (-14.692799048494292-7.425630747524895j), (12.828427124746192-7.242640687119286j), (-12.710780677203362-13.23154032980412j)]
numpy_result: [(41+0j), (-12.710780677203363-13.231540329804115j), (12.82842712474619-7.242640687119286j), (-14.692799048494292-7.4256307475248935j), (1+12j), (5.763866860359768-6.011417185151798j), (7.17157287525381-1.2426406871192857j), (-10.360287134662112-11.81732676743102j), (-3+0j), (-10.360287134662112+11.81732676743102j), (7.17157287525381+1.2426406871192857j), (5.763866860359768+6.011417185151798j), (1-12j), (-14.692799048494292+7.4256307475248935j), (12.82842712474619+7.242640687119286j), (-12.710780677203363+13.231540329804115j)]
difference: [0j, 26.46308065960823j, 14.485281374238571j, (-3.552713678800501e-15+14.851261495049787j), (1.3322676295501878e-15-24j), 12.022834370303597j, (-1.7763568394002505e-15+2.4852813742385695j), (-1.7763568394002505e-15+23.634653534862046j), 0j, -23.63465353486204j, -2.4852813742385704j, (3.552713678800501e-15-12.022834370303595j), (-1.3322676295501878e-15+24j), -14.851261495049789j, (1.7763568394002505e-15-14.485281374238571j), (1.7763568394002505e-15-26.463080659608238j)]
The IFFT result:
test IFFT
case1 [1, 1]
manual_result: [1.0, 0.0]
numpy_result: [(1+0j), 0j]
difference: [0j, 0j]
case2 [1, 2, 3, 4, 5, 6, 7, 8]
manual_result: [0.5625, (-0.0625+0.15088834764831843j), (-0.0625+0.062499999999999986j), (-0.0625+0.025888347648318405j), -0.0625, (-0.0625-0.025888347648318433j), (-0.0625-0.062499999999999986j), (-0.062499999999999986-0.1508883476483184j)]
numpy_result: [(4.5+0j), (-0.5-1.2071067811865475j), (-0.5-0.5j), (-0.5-0.20710678118654757j), (-0.5+0j), (-0.5+0.20710678118654757j), (-0.5+0.5j), (-0.5+1.2071067811865475j)]
difference: [(-3.9375+0j), (0.4375+1.357995128834866j), (0.4375+0.5625j), (0.4375+0.23299512883486598j), (0.4375+0j), (0.4375-0.232995128834866j), (0.4375-0.5625j), (0.4375-1.357995128834866j)]
case3 [1, 4, 2, 9, 0, 0, 3, 8, 9, 1, 4, 0, 0, 0, 0, 0]
manual_result: [0.0400390625, (-0.01241287175508141-0.012921426103324331j), (0.012527760864009951-0.007072891296014926j), (-0.014348436570795205-0.007251592526879778j), (0.0009765625000000013+0.01171875j), (0.005628776230820083-0.005870524594874804j), (0.007003489135990047-0.0012135162960149274j), (-0.01011746790494347-0.011540358171319353j), -0.0029296875, (-0.010117467904943469+0.011540358171319355j), (0.007003489135990049+0.0012135162960149274j), (0.005628776230820081+0.005870524594874803j), (0.0009765624999999987-0.01171875j), (-0.014348436570795205+0.0072515925268797805j), (0.012527760864009953+0.007072891296014926j), (-0.012412871755081408+0.01292142610332433j)]
numpy_result: [(2.5625+0j), (-0.7944237923252102+0.8269712706127572j), (0.8017766952966369+0.45266504294495535j), (-0.9182999405308933+0.46410192172030584j), (0.0625-0.75j), (0.3602416787724855+0.37571357407198736j), (0.44822330470336313+0.07766504294495535j), (-0.647517945916382+0.7385829229644387j), (-0.1875+0j), (-0.647517945916382-0.7385829229644387j), (0.44822330470336313-0.07766504294495535j), (0.3602416787724855-0.37571357407198736j), (0.0625+0.75j), (-0.9182999405308933-0.46410192172030584j), (0.8017766952966369-0.45266504294495535j), (-0.7944237923252102-0.8269712706127572j)]
difference: [(-2.5224609375+0j), (0.7820109205701288-0.8398926967160816j), (-0.7892489344326269-0.45973793424097026j), (0.903951503960098-0.47135351424718563j), (-0.0615234375+0.76171875j), (-0.3546129025416654-0.38158409866686216j), (-0.4412198155673731-0.07887855924097029j), (0.6374004780114385-0.7501232811357581j), (0.1845703125+0j), (0.6374004780114385+0.7501232811357581j), (-0.4412198155673731+0.07887855924097029j), (-0.3546129025416654+0.38158409866686216j), (-0.0615234375-0.76171875j), (0.903951503960098+0.47135351424718563j), (-0.7892489344326269+0.45973793424097026j), (0.7820109205701288+0.8398926967160816j)]
#pjs, Thank you for your reminder that FFT requires len(data) to be a power of 2.
As was pointed out in comments, you used a positive sign in the computation of omg_n. There are different definitions of the DFT, so it isn't wrong by itself. However this would naturally lead to differences if you compare your results with an implementation that uses a negative sign, as is the case with numpy.fft.fft. Adjusting your implementation to also use a negative sign would cover all forward transform cases (leaving only small roundoff errors on the order of ~10-16).
For the inverse transform cases, your implementation ends up scaling the result by 1/n at every stage, instead of only the final stage. To correct this, simply remove the scaling from the recursion, and normalize only on the final stage:
def FFTrecursion(a: List, flag: bool) -> List:
"""Recursion of the FFT implementation"""
n = len(a)
if n == 1:
return a
# complex root
omg_n = exp(-2 * pi * 1j / n)
if flag:
# IFFT
omg_n = 1 / omg_n
omg = 1
# split a into 2 part
a0 = a[::2] # even
a1 = a[1::2] # odd
# corresponding y
y0 = FFTrecursion(a0, flag)
y1 = FFTrecursion(a1, flag)
# result y
y = [0] * n
for k in range(n // 2):
y[k] = y0[k] + omg * y1[k]
y[k + n // 2] = y0[k] - omg * y1[k]
omg = omg * omg_n
return y
def FFT(a: List, flag: bool) -> List:
"""realize DFT using FFT"""
y = FFTrecursion(a, flag)
# IFFT final scaling
if flag:
n = len(a)
y = [i / n for i in y]
return y
Given boundary value k, is there a vectorized way to replace each number n with consecutive descending numbers from n-1 to k? For example, if k is 0 the I'd like to replace np.array([3,4,2,2,1,3,1]) with np.array([2,1,0,3,2,1,0,1,0,1,0,0,2,1,0,0]). Every item of input array is greater than k.
I have tried combination of np.repeat and np.cumsum but it seems evasive solution:
x = np.array([3,4,2,2,1,3,1])
y = np.repeat(x, x)
t = -np.ones(y.shape[0])
t[np.r_[0, np.cumsum(x)[:-1]]] = x-1
np.cumsum(t)
Is there any other way? I expect smth like inverse of np.add.reduceat that is able to broadcast integers to decreasing sequences instead of minimizing them.
Here's another way with array-assignment to skip the repeat part -
def func1(a):
l = a.sum()
out = np.full(l, -1, dtype=int)
out[0] = a[0]-1
idx = a.cumsum()[:-1]
out[idx] = a[1:]-1
return out.cumsum()
Benchmarking
# OP's soln
def OP(x):
y = np.repeat(x, x)
t = -np.ones(y.shape[0], dtype=int)
t[np.r_[0, np.cumsum(x)[:-1]]] = x-1
return np.cumsum(t)
Using benchit package (few benchmarking tools packaged together; disclaimer: I am its author) to benchmark proposed solutions.
import benchit
a = np.array([3,4,2,2,1,3,1])
in_ = [np.resize(a,n) for n in [10, 100, 1000, 10000]]
funcs = [OP, func1]
t = benchit.timings(funcs, in_)
t.plot(logx=True, save='timings.png')
Extend to take k as arg
def func1(a, k):
l = a.sum()+len(a)*(-k)
out = np.full(l, -1, dtype=int)
out[0] = a[0]-1
idx = (a-k).cumsum()[:-1]
out[idx] = a[1:]-1-k
return out.cumsum()
Sample run -
In [120]: a
Out[120]: array([3, 4, 2, 2, 1, 3, 1])
In [121]: func1(a, k=-1)
Out[121]:
array([ 2, 1, 0, -1, 3, 2, 1, 0, -1, 1, 0, -1, 1, 0, -1, 0, -1,
2, 1, 0, -1, 0, -1])
This is concise and probably ok for efficiency; I don't think apply is vectorized here, so you will be limited mostly be the number of elements in the original array (less so their value is my guess):
import pandas as pd
x = np.array([3,4,2,2,1,3,1])
values = pd.Series(x).apply(lambda val: np.arange(val-1,-1,-1)).values
output = np.concatenate(values)
I have a square 2D numpy array, A, and an array of zeros, B, with the same shape.
For every index (i, j) in A, other than the first and last rows and columns, I want to assign to B[i, j] the value of np.sum(A[i - 1:i + 2, j - 1:j + 2].
Example:
A =
array([[0, 0, 0, 0, 0],
[0, 1, 0, 1, 0],
[0, 1, 1, 0, 0],
[0, 1, 0, 1, 0],
[0, 0, 0, 0, 0])
B =
array([[0, 0, 0, 0, 0],
[0, 3, 4, 2, 0],
[0, 4, 6, 3, 0],
[0, 3, 4, 2, 0],
[0, 0, 0, 0, 0])
Is there an efficient way to do this? Or should I simply use a for loop?
There is a clever (read "borderline smartass") way to do this with np.lib.stride_tricks.as_strided. as_strided allows you to create views into your buffer that simulate windows by adding another dimension to the view. For example, if you had a 1D array like
>>> x = np.arange(10)
>>> np.lib.stride_tricks.as_strided(x, shape=(3, x.shape[0] - 2), strides=x.strides * 2)
array([[0, 1, 2, 3, 4, 5, 6, 7],
[1, 2, 3, 4, 5, 6, 7, 8],
[2, 3, 4, 5, 6, 7, 8, 9]])
Hopefully it is clear that you can just sum along axis=0 to get the sum of each size 3 window. There is no reason you couldn't extrend that to two or more dimensions. I've written the shape and index of the previous example in a way that suggests a solution:
A = np.array([[0, 0, 0, 0, 0],
[0, 1, 0, 1, 0],
[0, 1, 1, 0, 0],
[0, 1, 0, 1, 0],
[0, 0, 0, 0, 0]])
view = np.lib.stride_tricks.as_strided(A,
shape=(3, 3, A.shape[0] - 2, A.shape[1] - 2),
strides=A.strides * 2
)
B[1:-1, 1:-1] = view.sum(axis=(0, 1))
Summing along multiple axes simultaneously has been supported in np.sum since v1.7.0. For older versions of numpy, just sum repeatedly (twice) along axis=0.
Filling in the edges of B is left as an exercise for the reader (since it's not really part of the question).
As an aside, the solution here is a one-liner if you want it to be. Personally, I think anything with as_strided is already illegible enough, and doesn't need any further obfuscation. I'm not sure if a for loop is going to be bad enough performance-wise to justify this method in fact.
For future reference, here is a generic window-making function that can be used to solve this sort of problem:
def window_view(a, window=3):
"""
Create a (read-only) view into `a` that defines window dimensions.
The first ``a.ndim`` dimensions of the returned view will be sized according to `window`.
The remaining ``a.ndim`` dimensions will be the original dimensions of `a`, truncated by `window - 1`.
The result can be post-precessed by reducing the leading dimensions. For example, a multi-dimensional moving average could look something like ::
window_view(a, window).sum(axis=tuple(range(a.ndim))) / window**a.ndim
If the window size were different for each dimension (`window` were a sequence rather than a scalar), the normalization would be ``np.prod(window)`` instead of ``window**a.ndim``.
Parameters
-----------
a : array-like
The array to window into. Due to numpy dimension constraints, can not have > 16 dims.
window :
Either a scalar indicating the window size for all dimensions, or a sequence of length `a.ndim` providing one size for each dimension.
Return
------
view : numpy.ndarray
A read-only view into `a` whose leading dimensions represent the requested windows into `a`.
``view.ndim == 2 * a.ndim``.
"""
a = np.array(a, copy=False, subok=True)
window = np.array(window, copy=False, subok=False, dtype=np.int)
if window.size == 1:
window = np.full(a.ndim, window)
elif window.size == a.ndim:
window = window.ravel()
else:
raise ValueError('Number of window sizes must match number of array dimensions')
shape = np.concatenate((window, a.shape))
shape[a.ndim:] -= window - 1
strides = a.strides * 2
return np.lib.stride_tricks.as_strided(a, shake=shape, strides=strides)
I have found no 'simple' ways of doing this. But here are two ways:
Still involves a for loop
# Basically, get the sum for each location and then pad the result with 0's
B = [[np.sum(A[j-1:j+2,i-1:i+2]) for i in range(1,len(A)-1)] for j in range(1,len(A[0])-1)]
B = np.pad(B, ((1,1)), "constant", constant_values=(0))
Is longer but no for loops (this will be a lot more efficient on big arrays):
# Roll basically slides the array in the desired direction
A_right = np.roll(A, -1, 1)
A_left = np.roll(A, 1, 1)
A_top = np.roll(A, 1, 0)
A_bottom = np.roll(A, -1, 0)
A_bot_right = np.roll(A_bottom, -1, 1)
A_bot_left = np.roll(A_bottom, 1, 1)
A_top_right = np.roll(A_top, -1, 1)
A_top_left = np.roll(A_top, 1, 1)
# After doing that, you can just add all those arrays and these operations
# are handled better directly by numpy compared to when you use for loops
B = A_right + A_left + A_top + A_bottom + A_top_left + A_top_right + A_bot_left + A_bot_right + A
# You can then return the edges to 0 or whatever you like
B[0:len(B),0] = 0
B[0:len(B),len(B[0])-1] = 0
B[0,0:len(B)] = 0
B[len(B[0])-1,0:len(B)] = 0
You can just sum the 9 arrays that make up a block, each one being shifted by 1 w.r.t. the previous in either dimension. Using slice notation this can be done for the whole array A at once:
B = np.zeros_like(A)
B[1:-1, 1:-1] = sum(A[i:A.shape[0]-2+i, j:A.shape[1]-2+j]
for i in range(0, 3) for j in range(0, 3))
General version for arbitrary rectangular windows
def sliding_window_sum(a, size):
"""Compute the sum of elements of a rectangular sliding window over the input array.
Parameters
----------
a : array_like
Two-dimensional input array.
size : int or tuple of int
The size of the window in row and column dimension; if int then a quadratic window is used.
Returns
-------
array
Shape is ``(a.shape[0] - size[0] + 1, a.shape[1] - size[1] + 1)``.
"""
if isinstance(size, int):
size = (size, size)
m = a.shape[0] - size[0] + 1
n = a.shape[1] - size[1] + 1
return sum(A[i:m+i, j:n+j] for i in range(0, size[0]) for j in range(0, size[1]))
What is the easiest/fastest way to take a weighted sum of values in a numpy array?
Example: Solving the heat equation with the Euler method
length_l=10
time_l=10
u=zeros((length_l,length_l))# (x,y)
u[:, 0]=1
u[:,-1]=1
print(u)
def dStep(ALPHA=0.1):
for position,value in ndenumerate(u):
D2u= (u[position+(1,0)]-2*value+u[position+(-1, 0)])/(1**2) \
+(u[position+(0,1)]-2*value+u[position+( 0,-1)])/(1**2)
value+=ALPHA*D2u()
while True:
dStep()
print(u)
D2u should be the second central difference in two dimensions. This would work if I could add indexes like (1,4)+(1,3)=(2,7). Unfortunately, python adds them as (1,4)+(1,3)=(1,4,1,3).
Note that computing D2u is equivalent to taking a dot product with this kernel centered around the current position:
0, 1, 0
1,-4, 1
0, 1, 0
Can this be vectorised as a dot product?
I think you want something like:
import numpy as np
from scipy.ndimage import convolve
length_l = 10
time_l = 10
u = np.zeros((length_l, length_l))# (x,y)
u[:, 0] = 1
u[:, -1] = 1
alpha = .1
weights = np.array([[ 0, 1, 0],
[ 1, -4, 1],
[ 0, 1, 0]])
for i in range(5):
u += alpha * convolve(u, weights)
print(u)
You could reduce down a bit by doing:
weights = alpha * weights
weights[1, 1] = weights[1, 1] + 1
for i in range(5):
u = convolve(u, weights)
print(u)