Context:
I have a hydraulic erosion algorithm that needs to receive an array of droplet starting positions. I also already have a pattern replicating algorithm, so I only need a good pattern to replicate.
The Requirements:
I need an algorism that produces a set of n^2 entries in a set of format (x,y) or [index] that describe cells in an nxn grid (where n = 2^i where i is any positive integer).
(as a set it means that every cell is mentioned in exactly one entry)
The pattern [created by the algorism ] should contain zero to none clustering of "visited" cells at any stage.
The cell (0,0) is as close to (n-1,n-1) as to (1,1), this relates to the definition of clustering
Note
I was/am trying to find solutions through fractal-like patterns built through recursion, but at the time of writing this, my solution is a lookup table of a checkerboard pattern(list of black cells + list of white cells) (which is bad, but yields fewer artifacts than an ordered list)
C, C++, C#, Java implementations (if any) are preferred
You can use a linear congruential generator to create an even distribution across your n×n space. For example, if you have a 64×64 grid, using a stride of 47 will create the pattern on the left below. (Run on jsbin) The cells are visited from light to dark.
That pattern does not cluster, but it is rather uniform. It uses a simple row-wide transformation where
k = (k + 47) mod (n * n)
x = k mod n
y = k div n
You can add a bit of randomness by making k the index of a space-filling curve such as the Hilbert curve. This will yield the pattern on the right. (Run on jsbin)
You can see the code in the jsbin links.
I have solved the problem myself and just sharing my solution:
here are my outputs for the i between 0 and 3:
power: 0
ordering:
0
matrix visit order:
0
power: 1
ordering:
0 3 2 1
matrix visit order:
0 3
2 1
power: 2
ordering:
0 10 8 2 5 15 13 7 4 14 12 6 1 11 9 3
matrix visit order:
0 12 3 15
8 4 11 7
2 14 1 13
10 6 9 5
power: 3
ordering:
0 36 32 4 18 54 50 22 16 52 48 20 2 38 34 6
9 45 41 13 27 63 59 31 25 61 57 29 11 47 43 15
8 44 40 12 26 62 58 30 24 60 56 28 10 46 42 14
1 37 33 5 19 55 51 23 17 53 49 21 3 39 35 7
matrix visit order:
0 48 12 60 3 51 15 63
32 16 44 28 35 19 47 31
8 56 4 52 11 59 7 55
40 24 36 20 43 27 39 23
2 50 14 62 1 49 13 61
34 18 46 30 33 17 45 29
10 58 6 54 9 57 5 53
42 26 38 22 41 25 37 21
the code:
public static int[] GetPattern(int power, int maxReturnSize = int.MaxValue)
{
int sideLength = 1 << power;
int cellsNumber = sideLength * sideLength;
int[] ret = new int[cellsNumber];
for ( int i = 0 ; i < cellsNumber && i < maxReturnSize ; i++ ) {
// this loop's body can be used for per-request computation
int x = 0;
int y = 0;
for ( int p = power - 1 ; p >= 0 ; p-- ) {
int temp = (i >> (p * 2)) % 4; //2 bits of the index starting from the begining
int a = temp % 2; // the first bit
int b = temp >> 1; // the second bit
x += a << power - 1 - p;
y += (a ^ b) << power - 1 - p;// ^ is XOR
// 00=>(0,0), 01 =>(1,1) 10 =>(0,1) 11 =>(1,0) scaled to 2^p where 0<=p
}
//to index
int index = y * sideLength + x;
ret[i] = index;
}
return ret;
}
I do admit that somewhere along the way the values got transposed, but it does not matter because of how it works.
After doing some optimization I came up with this loop body:
int x = 0;
int y = 0;
for ( int p = 0 ; p < power ; p++ ) {
int temp = ( i >> ( p * 2 ) ) & 3;
int a = temp & 1;
int b = temp >> 1;
x = ( x << 1 ) | a;
y = ( y << 1 ) | ( a ^ b );
}
int index = y * sideLength + x;
(the code assumes that c# optimizer, IL2CPP, and CPP compiler will optimize variables temp, a, b out)
I am looking for an efficient algorithm for the following problem:
There is an array with values, i.e. (note that index 0 is omitted on purpose)
Index 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12
Value 17, 12, 5, 22, 3, 12, 6, 13, 7, 0, 2, 15
What I need to find is a subset of indices under these constraints:
The number of indices is constant (i.e. 3)
The sum of indices is constant (i.e. 20)
Each index may only appear once (so [2, 9, 9] is not a valid solution)
The sum of values is maximum.
For example if the subset length is 3 and the sum is 20, all possible solutions would be
Indices: [1, 7, 12] Sum of values: 17 + 6 + 15 = 38
Indices: [1, 8, 11] Sum of values: 17 + 13 + 2 = 32
Indices: [1, 9, 10] Sum of values: 17 + 7 + 0 = 24
Indices: [2, 6, 12] Sum of values: 12 + 12 + 15 = 39
Indices: [2, 7, 11] Sum of values: 12 + 6 + 2 = 20
Indices: [2, 8, 10] Sum of values: 12 + 13 + 0 = 25
Indices: [3, 5, 12] Sum of values: 5 + 3 + 15 = 23
Indices: [3, 6, 11] Sum of values: 5 + 12 + 2 = 19
Indices: [3, 7, 10] Sum of values: 5 + 6 + 0 = 11
Indices: [3, 8, 9] Sum of values: 5 + 13 + 7 = 25
Indices: [4, 5, 11] Sum of values: 22 + 3 + 2 = 27
Indices: [4, 6, 10] Sum of values: 22 + 12 + 0 = 34
Indices: [4, 7, 9] Sum of values: 22 + 6 + 7 = 35
Indices: [5, 6, 9] Sum of values: 3 + 12 + 7 = 22
Indices: [5, 7, 8] Sum of values: 3 + 6 + 13 = 22
of which [2, 6, 12] is the optimal solution because it has the maximum sum of values.
At the moment I run through all possible combinations using a slightly modified partition algorithm which grows exponentially as the sum of indices grows, so I wonder if there is any better way?
Solution O(I.S.K)
Let's do some naming first:
I is the greatest index (12 in your example)
S is the sum of values whose indices are selected (20 in your example)
K is the number of selected indices
V[] the array of values linked to the indices
maxsum(s, i, k) the maximal sum reachable by using k indices, all differents, whose value is less than or equal to i and whose sum is s.
Then you want to find maxsum(S, I, K)
Your problem exhibits some good properties:
optimal sub-structure
redundant sub-problems
For instance, when trying to compute maxsum(s, i, k) I can either not use index i, in which case the value is maxsum(s, i-1, k). Or I could use index i. In this case, I want to solve the sub-problem: what is the maximum sum reachable by indices less than or equal to i-1 and whose sum is s-i using k-1 such indices. This is the value: V[i] + maxsum(s-i, i-1, k-1).
As we want to reach the maximal sum we end up having: (Edit: corrected maxsum(s-i, i-1, k) to maxsum(s-i, i-1, k-1))
maxsum(s, i, k) = max{ maxsum(s, i-1, k) ; V[i] + maxsum(s-i, i-1, k-1) }
This is typical of a problem solvable by dynamic programming.
Here is an example C++ program solving the problem in O(I.S.K) (space and time).
We can improve the space complexity to O(I.S) at the price of a bigger time complexity: O(I.S.K²).
How to use the program
g++ -std=c++14 -g -Wall -O0 dp.cpp -o dp
./dp input.txt
Where input.txt is a file with the following format:
first line contains three integers: I S K
second line contains I integers, the values of the indices
Example run
---- K=1 ----
17 12 5 22 3 12 6 13 7 0 2 15
[ 1][ 2][ 3][ 4][ 5][ 6][ 7][ 8][ 9][10][11][12]
[ 1] 17 17 17 17 17 17 17 17 17 17 17 17
[ 2] 12 12 12 12 12 12 12 12 12 12 12
[ 3] 5 5 5 5 5 5 5 5 5 5
[ 4] 22 22 22 22 22 22 22 22 22
[ 5] 3 3 3 3 3 3 3 3
[ 6] 12 12 12 12 12 12 12
[ 7] 6 6 6 6 6 6
[ 8] 13 13 13 13 13
[ 9] 7 7 7 7
[10] 0 0 0
[11] 2 2
[12] 15
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
---- K=2 ----
17 12 5 22 3 12 6 13 7 0 2 15
[ 1][ 2][ 3][ 4][ 5][ 6][ 7][ 8][ 9][10][11][12]
[ 1]
[ 2] 12 12 12 12 12 12 12 12 12 12 12
[ 3] 29 29 29 29 29 29 29 29 29 29 29
[ 4] 22 22 22 22 22 22 22 22 22 22
[ 5] 17 39 39 39 39 39 39 39 39 39
[ 6] 34 34 34 34 34 34 34 34 34
[ 7] 27 27 29 29 29 29 29 29 29
[ 8] 8 24 24 24 24 24 24 24
[ 9] 25 25 25 30 30 30 30 30
[10] 34 34 34 34 34 34 34
[11] 15 28 28 28 28 28 28
[12] 9 35 35 35 35 35
[13] 18 18 29 29 29 32
[14] 25 25 25 25 27
[15] 19 19 19 24 24
[16] 13 13 13 37
[17] 20 20 20 20
[18] 13 13 27
[19] 7 15 21
[20] 9 28
---- K=3 ----
17 12 5 22 3 12 6 13 7 0 2 15
[ 1][ 2][ 3][ 4][ 5][ 6][ 7][ 8][ 9][10][11][12]
[ 1]
[ 2]
[ 3]
[ 4]
[ 5] 17 17 17 17 17 17 17 17 17 17
[ 6] 34 34 34 34 34 34 34 34 34 34
[ 7] 51 51 51 51 51 51 51 51 51
[ 8] 44 44 44 44 44 44 44 44 44
[ 9] 39 39 41 41 41 41 41 41 41
[10] 42 42 42 42 42 42 42 42
[11] 37 51 51 51 51 51 51 51
[12] 30 46 46 46 46 46 46 46
[13] 39 40 52 52 52 52 52
[14] 20 35 47 47 47 47 47
[15] 37 37 42 42 42 42 44
[16] 31 37 37 37 41 41
[17] 40 40 40 40 40 54
[18] 21 47 47 47 47 49
[19] 41 41 41 41 44
[20] 22 35 35 35 39
index: 12 sum: 20
index: 6 sum: 8
index: 2 sum: 2
max sum: 39
The source code
#include <cstdio>
#include <iomanip>
#include <iostream>
#include <limits>
#include <valarray>
#include <vector>
using namespace std;
auto const INF = numeric_limits<double>::infinity();
struct matrix {
matrix(size_t rows, size_t cols, double value)
: cells(value, rows*cols)
, rows(rows)
, cols(cols)
, value(value)
{}
double& operator() (int r, int c)
{
if(r < 0 || c < 0)
return value;
return cells[r*cols+c];
}
valarray<double> cells;
size_t rows;
size_t cols;
double value;
};
int main(int argc, char* argv[]) {
if(argc > 1)
freopen(argv[1], "r", stdin);
// I: max index
// S: sum of indices
// K: number of indices in the sum S
int I, S, K;
cin >> I >> S >> K;
// load values
vector<double> V(I+1, 0);
for(int i=1; i<=I; ++i)
cin >> V[i];
// dynamic programming:
// --------------------
// maxsum(i, s, k) is the maximal sum reachable using 'k' indices, less
// than or equal to 'i', all differents, and having a sum of 's'
//
// maxsum(i, s, k) =
// -oo if i > s
//
// -oo if i < s && k == 1
//
// V[s] if i >= s && s <= I && k == 1
// -oo if (i < s || s > I) && k == 1
//
// max { V[i] + maxsum(i-1, S-i, k-1), maxsum(i-1, S, k) }
vector<matrix> maxsum(K+1, matrix(S+1, I+1, -INF));
// initialize K=1
for(int s=0; s<=I && s<=S; ++s) {
for(int i=s; i<=I; ++i) {
maxsum[1](s, i) = V[s];
}
}
// K > 1
for(int k=2; k<=K; ++k) {
for(int s=2; s<=S; ++s) {
for(int i=1; i<=I; ++i) {
auto l = V[i] + maxsum[k-1](s-i, i-1);
auto r = maxsum[k](s, i-1);
maxsum[k](s, i) = max(l, r);
}
}
}
// display the whole dynamic programming tables (optional)
for(int k=1; k<=K; ++k) {
cout << "---- K=" << k << " ----\n";
cout << " ";
for(int i=1; i<=I; ++i) {
cout << setw(3) << V[i] << ' ';
}
cout << '\n';
cout << " ";
for(int i=1; i<=I; ++i) {
cout << '[' << setw(2) << i << ']';
}
cout << '\n';
for(int s=1; s<=S; ++s) {
cout << '[' << setw(2) << s << "] ";
for(int i=1; i<=I; ++i) {
if(maxsum[k](s, i) == -INF) {
cout << " ";
} else {
cout << setw(3) << maxsum[k](s, i) << ' ';
}
}
cout << '\n';
}
}
// output the indices belonging to the solution by working backward in the
// dynamic programming tables
int t_S = S;
int t_I = I;
for(int k=K; k>=1; --k) {
if(t_I <= 0 || t_S <= 0) {
cout << "error...\n";
break;
}
auto m = maxsum[k](t_S, t_I);
int i;
for(i=t_I; i>=1; --i) {
if(maxsum[k](t_S, i) != m)
break;
}
cout << "index: " << setw(3) << (i+1) << ' ';
cout << "sum: " << setw(3) << t_S << '\n';
t_I = i;
t_S = t_S - i - 1;
}
cout << "max sum: " << maxsum[K](S, I) << '\n';
}
Take the arrays, and sort them by value instead of by index (keeping the index-value pairs preserved). Now, starting at the end of the array, take the last k numbers in the indices array, where k is the number of indices you have to have, and sum them up. If it equals the desired sum, great- you are done. If not, take note of the difference (desired sum - actual sum), and add that to the (n - k)th index. Find that index in the index array (ordered by value, mind you), now find your new sum of values (you can optimize this by subtracting out the old index's value and add the new one, instead of recomputing the sum of all k values).
You now have one valid solution, and a lower bound. You know the indices of the rest of the valid solution that can even possibly beat this score must come after the smallest index's value in the value-sorted array. That is:
Both sorted by value-
indices: | bunch of indices | index we found | more | k-1 'random' indices |
values: | bunch of values | value for ^ | more | k-1 largest values |
So we only have to search 'more' and the k-1 largest values for valid indices that satisfy the criteria and also have values that form a larger sum. To do this, we rinse and repeat, moving the smallest of the (n-k-1) elements backwards one, so we effectively try all combination of these elements, but in the order of decreasing subset-sums of our set of k elements. This allows us to continually narrow the space we search as we find larger sums, because know for certain that any sum that contains a smaller value than that of best solution will have a smaller sum (because the rest of the set is already as large as possible).
Pseudo Code:
pair_array = input() // each pair consists of index and value
sort_by_value(pair_array)
best_sum = 0
markers = [n - (k-1) .. n] // mark the k-1 indices being summed
while True:
sum_of_indices = sum_indices(pair_array[markers])
value_sum = sum_value(pair_array[markers])
if pair_array.contains(desired_sum - sum_of_indices): // this lets us effectively reduce our search by a factor of N, given contains uses a hashtable
value_sum += pair_array(pair_array.index(desired_sum - sum_of_indices)).value
if value_sum > best_sum:
best_sum = value_sum
pair_array.remove(0 .. n - (k-1)) // this greatly reduces the combinations checked
if has_next_combination(markers, pair_array):
next_greatest_combination(markers, pair_array) // pick new markers, in a reverse-binary counting fashion (most significant bit first way)
else:
print(best_sum)
break
One small trick I can think if you try to find lets say 3 indexes, is instead of iterating for the 3 indexes you can calculate the 3rd index when you know the first two indexes. For example when you know that
p1 = 1, p2 = 7 => p3 = 20 - (p1 + p2) = 12
This can be generalized when having N indexes the last one can always be inferred from the N-1 previous indexes.
I tried this in Python:
Index = [ 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12 ]
Value = [17, 12, 5, 22, 3, 12, 6, 13, 7, 0, 2, 15 ]
maxSum = 0 SumInd = 20
for p1 in range(1,len(Index)+1):
for p2 in range(p1,len(Index)+1):
if (p1 + p2 < SumInd) and (p1 != p2):
p3 = SumInd - (p1 + p2) #If you know p1 and p2 calculate p3 index
if (p3 < len(Index)+1) and (p2 != p3) and (p1 != p3):
fooSum = Value[p1-1]+Value[p2-1] + Value[p3-1]
print(p1,p2,p3 , "Sum is ",fooSum)
if maxSum < fooSum:
maxSum = fooSum
print("Max Sum is ", maxSum)
Of course you need to save the indexes when you find maxSum. Also this implementation calculate many similar pairs (e.g. [1,9,10] and [9,10,1]). Perhaps a better solution could eliminate this pairs.
EDIT: Big Improvement I found a way to eliminate most unnecessary checks. Let's say you need 3 indexes. The first one checks the whole range of possible values. Let's say it is index1.
The other two indexes must sum at 20 - ind1 let's call this rest. The list of indexes is always ordered so you can have a back index showing the smallest value(first item in the list bigger than index1) and a front index that show the biggest values (last item in the list). So index2 = backIndex , index3 = frontIndex.
If the rest is smaller than the sum of index2,3 you could increase the back index (get the next bigger value) or if it is larger you decrease the front index until the two indexes meet each other where you break and increase index1. This eliminates checking (1,7,12) and (1,12,7) two times.
The code is here in Python:
maxSum = 0
SumInd = 20
for index_1 in range(1,len(Index)):
rest = SumInd - index_1
backIndex = index_1+1
frontIndex = len(Index)
while backIndex < frontIndex:
if rest > (backIndex + frontIndex):
backIndex = backIndex + 1
elif rest < (backIndex + frontIndex):
frontIndex = frontIndex - 1
else:
fooSum = Value[index_1-1]+Value[backIndex-1] + Value[frontIndex-1]
print("Checking for ",index_1,backIndex,frontIndex,' Sum of values:',fooSum)
if maxSum < fooSum:
indList = [index_1-1,backIndex,frontIndex]
maxSum = fooSum
backIndex = backIndex + 1 #To avoid Inf loop
print("Max Sum is ", maxSum,"at",indList)
and gives these results:
Checking for 1 7 12 Sum of values: 38
Checking for 1 8 11 Sum of values: 32
Checking for 1 9 10 Sum of values: 24
Checking for 2 6 12 Sum of values: 39
Checking for 2 7 11 Sum of values: 20
Checking for 2 8 10 Sum of values: 25
Checking for 3 5 12 Sum of values: 23
Checking for 3 6 11 Sum of values: 19
Checking for 3 7 10 Sum of values: 11
Checking for 3 8 9 Sum of values: 25
Checking for 4 5 11 Sum of values: 27
Checking for 4 6 10 Sum of values: 34
Checking for 4 7 9 Sum of values: 35
Checking for 5 6 9 Sum of values: 22
Checking for 5 7 8 Sum of values: 22
Max Sum is 39 at [1, 6, 12]
This can always be generalized for N indexes. The first N-2 indexes can search the whole range of the list (like index 1 in the case above, also it should be noted that all these indexes start checking from previous index value plus one until the end of the list to eliminate many duplicate checks).
The last two indexes can be calculated like I showed in my code and avoid many duplicate checks.
Consider we have N points on a circle. To each point an index is assigned i = (1,2,...,N). Now, for a randomly selected point, I want to have a vector including the indices of 5 points, [two left neighbors, the point itself, two right neighbors].
See the figure below.
Some sxamples are as follows:
N = 18;
selectedPointIdx = 4;
sequence = [2 3 4 5 6];
selectedPointIdx = 1
sequence = [17 18 1 2 3]
selectedPointIdx = 17
sequence = [15 16 17 18 1];
The conventional way to code this is considering the exceptions as if-else statements, as I did:
if ii == 1
lseq = [N-1 N ii ii+1 ii+2];
elseif ii == 2
lseq = [N ii-1 ii ii+1 ii+2];
elseif ii == N-1
lseq=[ii-2 ii-1 ii N 1];
elseif ii == N
lseq=[ii-2 ii-1 ii 1 2];
else
lseq=[ii-2 ii-1 ii ii+1 ii+2];
end
where ii is selectedPointIdx.
It is not efficient if I consider for instance 7 points instead of 5. What is a more efficient way?
How about this -
off = -2:2
out = mod((off + selectedPointIdx) + 17,18) + 1
For a window size of 7, edit off to -3:3.
It uses the strategy of subtracting 1 + modding + adding back 1 as also discussed here.
Sample run -
>> off = -2:2;
for selectedPointIdx = 1:18
disp(['For selectedPointIdx =',num2str(selectedPointIdx),' :'])
disp(mod((off + selectedPointIdx) + 17,18) + 1)
end
For selectedPointIdx =1 :
17 18 1 2 3
For selectedPointIdx =2 :
18 1 2 3 4
For selectedPointIdx =3 :
1 2 3 4 5
For selectedPointIdx =4 :
2 3 4 5 6
For selectedPointIdx =5 :
3 4 5 6 7
For selectedPointIdx =6 :
4 5 6 7 8
....
For selectedPointIdx =11 :
9 10 11 12 13
For selectedPointIdx =12 :
10 11 12 13 14
For selectedPointIdx =13 :
11 12 13 14 15
For selectedPointIdx =14 :
12 13 14 15 16
For selectedPointIdx =15 :
13 14 15 16 17
For selectedPointIdx =16 :
14 15 16 17 18
For selectedPointIdx =17 :
15 16 17 18 1
For selectedPointIdx =18 :
16 17 18 1 2
You can use modular arithmetic instead: Let p be the point among N points numbered 1 to N. Say you want m neighbors on each side, you can get them as follows:
(p - m - 1) mod N + 1
...
(p - 4) mod N + 1
(p - 3) mod N + 1
(p - 2) mod N + 1
p
(p + 1) mod N + 1
(p + 2) mod N + 1
(p + 3) mod N + 1
...
(p + m - 1) mod N + 1
Code:
N = 18;
p = 2;
m = 3;
for i = p - m : p + m
nb = mod((i - 1) , N) + 1;
disp(nb);
end
Run code here
I would like you to note that you might not necessarily improve performance by avoiding a if statement. A benchmark might be necessary to figure this out. However, this will only be significant if you are treating tens of thousands of numbers.