I think this is a very complicated dynamic programming problem.
Two spies each have a secret number in [1..m]. To exchange numbers they agree to meet at the river and "innocently" take turns throwing stones: from a pile of n=26 identical stones, each spy in turn throws at least one stone in the river.
The only information is in the number of stones each thrown in each turn. What is the largest m can be so they are sure they can complete the exchange?
Develop a recursive formula to count. Here is the start of the table; complete it to n=26. (You should not expect a closed form.)
n 1 2 3 4 5 6 7 8 9 10 11 12
m 1 1 1 2 2 3 4 6 8 12 16 23
Here are some hints from our professor: I suggest changing the problem to making the following table: Let R(n,m) be the range of numbers [1..R(n,m)] that A can indicate to B if they start with n stones, and both know that A has to also receive a number in [1..m] from B.
For example, if A needs no more information, R(n,1) can be computed by considering how many stones A could throw (one to n), then B thows 1 (if any remain) and A gets to decide again. The base cases R(0,1) = R(1,1) = 1, and you can write a recursive rule if you are careful at the boundaries. (You should find the Fibonacci numbers for R(n,1).)
If A needs information, then B has to send it by his or her choices, so things are a little more complicated. Here is the start of the table:
n\ m 1 2 3 4 5
0 1 0 0 0 0
1 1 0 0 0 0
2 2 0 0 0 0
3 3 1 0 0 0
4 5 2 1 0 0
5 8 4 2 1 1
6 13 7 4 3 2
7 21 12 8 6 4
8 34 20 15 11 8
9 55 33 27 19 16
From the R(n,m) table, how would you recover the entries of the earlier table (the table showing m as a function of n)?
Related
I have got a sequence 1 2 3 4 5 6 ... n. Now, I am given a sequence of n deletions - each deletion is a number which I want to delete. I need to respond to each deletion with two numbers - of a left and right neighbour of deleted number (-1 if any doesn't exists).
E.g. I delete 2 - I respond 1 3, then I delete 3 I respond 1 4 , I delete 6 I respond 5 -1 etc.
I want to do it fast - linear of linear-logarithmic time complexity.
What data structure should I use? I guess the key to the solution is the fact that the sequence is sorted.
A doubly-linked list will do fine.
We will store the links in two arrays, prev and next, to allow O(1) access for deletions.
First, for every element and two sentinels at the ends, link it to the previous and next integers:
init ():
for cur := 0, 1, 2, ..., n, n+1:
prev[cur] := cur-1
next[cur] := cur+1
When you delete an element cur, update the links in O(1) like this:
remove (cur):
print (num (prev[cur]), " ", num (next[cur]), newline)
prev[next[cur]] := prev[cur]
next[prev[cur]] := next[cur]
Here, the num wrapper is inserted to print -1 for the sentinels:
num (cur):
if (cur == 0) or (cur == n+1):
return -1
else:
return cur
Here's how it works:
prev next
n = 6 prev/ print 0 1 2 3 4 5 6 7 0 1 2 3 4 5 6 7
/next ------------------- -------------------
init () -1 0 1 2 3 4 5 6 1 2 3 4 5 6 7 8
remove (2) 1 3 1 3 -1 0 1 3 4 5 6 1 3 4 5 6 7 8
remove (3) 1 4 1 4 -1 0 1 4 5 6 1 4 5 6 7 8
remove (6) 5 7 5 -1 -1 0 1 4 5 1 4 5 7 8
remove (1) 0 4 -1 4 -1 0 4 5 4 5 7 8
remove (5) 4 7 4 -1 -1 0 4 4 7 8
remove (4) 0 7 -1 -1 -1 0 7 8
Above, the portions not used anymore are blanked out for clarity.
The respective elements of the arrays still store the values printed above them, but we no longer access them.
As Jim Mischel rightly noted (thanks!), storing the list in two arrays instead of dynamically allocating the storage is crucial to make this O(1) per deletion.
You can use a binary search tree. Deleting from it is logarithmic. If you want to remove n elements and the number of total elements is m, then the complexity of removing n elements from it will be
nlogm
I feel like my run time is extremely slow for my data set, this is the code:
library(caret)
library(data.table)
knnImputeValues <- preProcess(mainData[trainingRows, imputeColumns], method = c("zv", "knnImpute"))
knnTransformed <- predict(knnImputeValues, mainData[ 1:1000, imputeColumns])
the PreProcess into knnImputeValues run's fairly quickly, however the predict function takes a tremendous amount of time. When I calculated it on a subset of the data this was the result:
testtime <- system.time(knnTransformed <- predict(knnImputeValues, mainData[ 1:15000, imputeColumns
testtime
user 969.78
system 38.70
elapsed 1010.72
Additionally, it should be noted that caret preprocess uses "RANN".
Now my full dataset is:
str(mainData[ , imputeColumns])
'data.frame': 1809032 obs. of 16 variables:
$ V1: int 3 5 5 4 4 4 3 4 3 3 ...
$ V2: Factor w/ 3 levels "1000000","1500000",..: 1 1 3 1 1 1 1 3 1 1 ...
$ V3: Factor w/ 2 levels "0","1": 2 2 2 2 2 2 2 2 2 2 ...
$ V4: int 2 5 5 12 4 5 11 8 7 8 ...
$ V5: int 2 0 0 2 0 0 1 3 2 8 ...
$ V6: int 648 489 489 472 472 472 497 642 696 696 ...
$ V7: Factor w/ 4 levels "","N","U","Y": 4 1 1 1 1 1 1 1 1 1 ...
$ V8: int 0 0 0 0 0 0 0 1 1 1 ...
$ V9: num 0 0 0 0 0 ...
$ V10: Factor w/ 56 levels "1","2","3","4",..: 45 19 19 19 19 19 19 46 46 46 ...
$ V11: Factor w/ 2 levels "0","1": 2 2 2 2 2 2 2 2 2 2 ...
$ V12: num 2 5 5 12 4 5 11 8 7 8 ...
$ V13: num 2 0 0 2 0 0 1 3 2 8 ...
$ V14: Factor w/ 4 levels "1","2","3","4": 2 2 2 2 2 2 2 2 3 3 ...
$ V15: Factor w/ 2 levels "0","1": 1 1 1 1 1 1 1 2 2 2 ...
$ V16: num 657 756 756 756 756 ...
So is there something I'm doing wrong, or is this typical for how long it will take to run this? If you back of the envelop extrapolate (which I know isn't entire accurate) you'd get what 33 days?
Also it looks like system time is very low and user time is very high, is that normal?
My computer is a laptop, with a Intel(R) Core(TM) i5-6300U CPU # 2.40Ghz processor.
Additionally would this improve the runtime of the predict function?
cl <- makeCluster(4)
registerDoParallel()
I tried it, and it didn't seem to make a difference other than all the processors looked more active in my task manager.
FOCUSED QUESTION: I'm using Caret package to do KNN Imputation on 1.8 Million Rows, the way I'm currently doing it will take over a month to run, how do I write this in such a way that I could do it in a much faster amount of time(if possible)?
Thank you for any help provided. And the answer might very well be "that's how long it takes don't bother" I just want to rule out any possible mistakes.
You can speed this up via the imputation package and use of canopies which can be installed from Github:
Sys.setenv("PKG_CXXFLAGS"="-std=c++0x")
devtools::install_github("alexwhitworth/imputation")
Canopies use a cheap distance metric--in this case distance from the data mean vector--to get approximate neighbors. In general, we wish to keep the canopies each sized < 100k so for 1.8M rows, we'll use 20 canopies:
library("imputation")
to_impute <- mainData[trainingRows, imputeColumns] ## OP undefined
imputed <- kNN_impute(to_impute, k= 10, q= 2, verbose= TRUE,
parallel= TRUE, n_canopies= 20)
NOTE:
The imputation package requires numeric data inputs. You have several factor variables in your str output. They will cause this to fail.
You'll also get some mean vector imputation if you have fulling missing rows.
# note this example data is too small for canopies to be useful
# meant solely to illustrate
set.seed(2143L)
x1 <- matrix(rnorm(1000), 100, 10)
x1[sample(1:1000, size= 50, replace= FALSE)] <- NA
x_imp <- kNN_impute(x1, k=5, q=2, n_canopies= 10)
sum(is.na(x_imp[[1]])) # 0
# with fully missing rows
x2 <- x1; x2[5,] <- NA
x_imp <- kNN_impute(x2, k=5, q=2, n_canopies= 10)
[1] "Computing canopies kNN solution provided within canopies"
[1] "Canopies complete... calculating kNN."
row(s) 1 are entirely missing.
These row(s)' values will be imputed to column means.
Warning message:
In FUN(X[[i]], ...) :
Rows with entirely missing values imputed to column means.
A matrix of size nxn needs to be constructed with the desired properties.
n is even. (given as input to the algorithm)
Matrix should contain integers from 0 to n-1
Main diagonal should contain only zeroes and matrix should be symmetric.
All numbers in each row should be different.
For various n , any one of the possible output is required.
input
2
output
0 1
1 0
input
4
output
0 1 3 2
1 0 2 3
3 2 0 1
2 3 1 0
Now the only idea that comes to my mind is to brute-force build combinations recursively and prune.
How can this be done in a iterative way perhaps efficiently?
IMO, You can handle your answer by an algorithm to handle this:
If 8x8 result is:
0 1 2 3 4 5 6 7
1 0 3 2 5 4 7 6
2 3 0 1 6 7 4 5
3 2 1 0 7 6 5 4
4 5 6 7 0 1 2 3
5 4 7 6 1 0 3 2
6 7 4 5 2 3 0 1
7 6 5 4 3 2 1 0
You have actually a matrix of two 4x4 matrices in below pattern:
m0 => 0 1 2 3 m1 => 4 5 6 7 pattern => m0 m1
1 0 3 2 5 4 7 6 m1 m0
2 3 0 1 6 7 4 5
3 2 1 0 7 6 5 4
And also each 4x4 is a matrix of two 2x2 matrices with a relation to a power of 2:
m0 => 0 1 m1 => 2 3 pattern => m0 m1
1 0 3 2 m1 m0
In other explanation I should say you have a 2x2 matrix of 0 and 1 then you expand it to a 4x4 matrix by replacing each cell with a new 2x2 matrix:
0 => 0+2*0 1+2*0 1=> 0+2*1 1+2*1
1+2*0 0+2*0 1+2*1 0+2*1
result => 0 1 2 3
1 0 3 2
2 3 0 1
3 2 1 0
Now expand it again:
0,1=> as above 2=> 0+2*2 1+2*2 3=> 0+2*3 1+2*3
1+2*2 0+2*2 1+2*3 0+2*3
I can calculate value of each cell by this C# sample code:
// i: row, j: column, n: matrix dimension
var v = 0;
var m = 2;
do
{
var p = m/2;
v = v*2 + (i%(n/p) < n/m == j%(n/p) < n/m ? 0 : 1);
m *= 2;
} while (m <= n);
We know each row must contain each number. Likewise, each row contains each number.
Let us take CS convention of indices starting from 0.
First, consider how to place the 1's in the matrix. Choose a random number k0, from 1 to n-1. Place the 1 in row 0 at position (0,k0). In row 1, if k0 = 1 in which case there is already a one placed. Otherwise, there are n-2 free positions and place the 1 at position (1,k1). Continue in this way until all the 1 are placed. In the final row there is exactly one free position.
Next, repeat with the 2 which have to fit in the remaining places.
Now the problem is that we might not be able to actually complete the square. We may find there are some constraints which make it impossible to fill in the last digits. The problem is that checking a partially filled latin square is NP-complete.(wikipedia) This basically means pretty compute intensive and there no know short-cut algorithm. So I think the best you can do is generate squares and test if they work or not.
If you only want one particular square for each n then there might be simpler ways of generating them.
The link Ted Hopp gave in his comment Latin Squares. Simple Construction does provide a method for generating a square starting with the addition of integers mod n.
I might be wrong, but if you just look for printing a symmetric table - a special case of latin squares isomorphic to the symmetric difference operation table over a powerset({0,1,..,n}) mapped to a ring {0,1,2,..,2^n-1}.
One can also produce such a table, using XOR(i,j) where i and j are n*n table indexes.
For example:
def latin_powerset(n):
for i in range(n):
for j in range(n):
yield (i, j, i^j)
Printing tuples coming from previously defined special-case generator of symmetric latin squares declared above:
def print_latin_square(sq, n=None):
cells = [c for c in sq]
if n is None:
# find the length of the square side
n = 1; n2 = len(cells)
while n2 != n*n:
n += 1
rows = list()
for i in range(n):
rows.append(" ".join("{0}".format(cells[i*n + j][2]) for j in range(n)))
print("\n".join(rows))
square = latin_powerset(8)
print(print_latin_square(square))
outputs:
0 1 2 3 4 5 6 7
1 0 3 2 5 4 7 6
2 3 0 1 6 7 4 5
3 2 1 0 7 6 5 4
4 5 6 7 0 1 2 3
5 4 7 6 1 0 3 2
6 7 4 5 2 3 0 1
7 6 5 4 3 2 1 0
See also
This covers more generic cases of latin squares, rather than that super symmetrical case with the trivial code above:
https://www.cut-the-knot.org/arithmetic/latin2.shtml (also pointed in the comments above for symmetric latin square construction)
https://doc.sagemath.org/html/en/reference/combinat/sage/combinat/matrices/latin.html
The third item in the FinnAPL Library is called “Cumulative maxima (⌈) of subvectors of Y indicated by X ” where X is a binary vector and Y os a vector of numbers. Here's an example of its usage:
X←1 0 0 0 1 0 0 0
Y←9 78 3 2 50 7 69 22
Y[A⍳⌈\A←⍋A[⍋(+\X)[A←⍋Y]]] ⍝ output 9 78 78 78 50 50 69 69
You can see that beginning from either the beginning or from any 1 value in the X array, the cumulave maximum is found for all corresponding digits in Y until another 1 is found in X. In the example given, X is divding the array into two equal parts of 4 numbers each. In the first part, 9 is the maxima until 78 is encountered, and in the second part 50 is the maxima until 69 is encountered.
That's easy enough to understand, and I could blindly use it as is, but I'd like to understand how it works, because APL idioms are essentially algorithms made up of operators and functions. To understand APL well, it's important to understand how the masters were able to weave it all together into such compact and elegant lines of code.
I find this particular idiom especially hard to understand because of the indexing nested two layers deep. So my question is, what makes this idiom tick?
This idiom can be broken down into smaller idioms, and most importantly, it contains idiom #11 from the FinnAPL Library entitled:
Grade up (⍋) for sorting subvectors of Y indicated by X
Using the same values for X and Y given in the question, here's an example of its usage:
X←1 0 0 0 1 0 0 0
Y←9 78 3 2 50 7 69 22
A[⍋(+\X)[A←⍋Y]] ⍝ output 4 3 1 2 6 8 5 7
As before, X is dividing the vector into two halves, and the output indicates, for each position, what digit of Y is needed to sort each of the halves. So, the 4 in the output is saying that it needs the 4th digit of Y (2) in the 1st position; the 3 indicates the 3rd digit (3) in the 2nd position; the 1 indicates the 1st digit (9) in the third position; etc. Thus, if we apply this indexing to Y, we get:
Y[A[⍋(+\X)[A←⍋Y]]] ⍝ output 2 3 9 78 7 22 50 69
In order to understand the indexing within this grade-up idiom, consider what is happening with the following:
(+\X)[A←⍋Y] ⍝ Sorted Cumulative Addition
Breaking it down step by step:
A←⍋Y ⍝ 4 3 6 1 8 5 7 2
+\X ⍝ 1 1 1 1 2 2 2 2
(+\X)[A←⍋Y] ⍝ 1 1 2 1 2 2 2 1 SCA
A[⍋(+\X)[A←⍋Y]] ⍝ 4 3 1 2 6 8 5 7
You can see that sorted cumulative addition (SCA) of X 1 1 2 1 2 2 2 1 applied to A acts as a combination of compress left and compress right. All values of A that line up with a 1 are moved to the left, and those lining up with a 2 move to the right. Of course, if X had more 1s, it would be compressing and locating the compressed packets in the order indicated by the values of the SCA result. For example, if the SCA of X were like 3 3 2 1 2 2 1 1 1, you would end up with the 4 digits corresponding to the 1s, followed by the 3 digits corresponding to the 2s, and finally, the 2 digits corresponding to the 3s.
You may have noticed that I skipped the step that would show the effect of grade up ⍋:
(+\X)[A←⍋Y] ⍝ 1 1 2 1 2 2 2 1 SCA
⍋(+\X)[A←⍋Y] ⍝ 1 2 4 8 3 5 6 7 Grade up
A[⍋(+\X)[A←⍋Y]] ⍝ 4 3 1 2 6 8 5 7
The effect of compression and rearrangement isn't accomplised by SCA alone. It effectively acts as rank, as I discussed in another post. Also in that post, I talked about how rank and index are essentially two sides of the same coin, and you can use grade up to switch between the two. Therefore, that is what is happening here: SCA is being converted to an index to apply to A, and the effect is grade-up sorted subvectors as indicated by X.
From Sorted Subvectors to Cumulative Maxima
As already described, the result of sorting the subvectors is an index, which when applied to Y, compresses the data into packets and arranges those packets according to X. The point is that it is an index, and once again, grade up is applied, which converts indexes into ranks:
⍋A[⍋(+\X)[A←⍋Y]] ⍝ 3 4 2 1 7 5 8 6
The question here is, why? Well, the next step is applying a cumulative maxima, and that really only makes sense if it is applied to values for rank which represent relative magnitude within each packet. Looking at the values, you can see that 4 is is the maxima for the first group of 4, and 8 is for the second group. Those values correspond to the input values of 78 and 69, which is what we want. It doesn't make sense (at least in this case) to apply a maxima to index values, which represent position, so the conversion to rank is necessary. Applying the cumulative maxima gives:
⌈\A←⍋A[⍋(+\X)[A←⍋Y]] ⍝ 3 4 4 4 7 7 8 8
That leaves one last step to finish the index. After doing a cumulative maxima operation, the vector values still represent rank, so they need to be converted back to index values. To do that, the index-of operator is used. It takes the value in the right argument and returns their position as found in the left argument:
A⍳⌈\A←⍋A[⍋(+\X)[A←⍋Y]] ⍝ 1 2 2 2 5 5 7 7
To make it easier to see:
3 4 2 1 7 5 8 6 left argument
3 4 4 4 7 7 8 8 right argument
1 2 2 2 5 5 7 7 result
The 4 is in the 2nd position in the left argument, so the result shows a 2 for every 4 in the right argument. The index is complete, so applying it to Y, we get the expected result:
Y[A⍳⌈\A←⍋A[⍋(+\X)[A←⍋Y]]] ⍝ 9 78 78 78 50 50 69 69
My implementation:
X←1 0 0 0 1 0 0 0
Y←9 78 3 2 50 7 69 22
¯1+X/⍳⍴X ⍝ position
0 4
(,¨¯1+X/⍳⍴X)↓¨⊂Y
9 78 3 2 50 7 69 22 50 7 69 22
(1↓(X,1)/⍳⍴X,1)-X/⍳⍴X ⍝ length
4 4
(,¨(1↓(X,1)/⍳⍴X,1)-X/⍳⍴X)↑¨(,¨¯1+X/⍳⍴X)↓¨⊂Y
9 78 3 2 50 7 69 22
⌈\¨(,¨(1↓(X,1)/⍳⍴X,1)-X/⍳⍴X)↑¨(,¨¯1+X/⍳⍴X)↓¨⊂Y
9 78 78 78 50 50 69 69
∊⌈\¨(,¨(1↓(X,1)/⍳⍴X,1)-X/⍳⍴X)↑¨(,¨¯1+X/⍳⍴X)↓¨⊂Y
9 78 78 78 50 50 69 69
Have a nice day.
I was asked to find a 11x11-grid containing the digits such that one can read the squares of 1,...,100. Here read means that you fix the starting position and direction (8 possibilities) and if you can find for example the digits 1,0,0,0,0,4 consecutively, you have found the squares of 1, 2, 10, 100 and 20. I made a program (the algorithm is not my own. I modified slightly a program which uses best-first search to find a solution but it is too slow. Does anyone know a better algorithm to solve the problem?
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <time.h>
#include <vector>
#include <algorithm>
using namespace std;
int val[21][21];//number which is present on position
int vnum[21][21];//number of times the position is used - useful if you want to backtrack
//5 unit borders
int mx[4]={-1,0,1,0};//movement arrays
int my[4]={0,-1,0,1};
int check(int x,int y,int v,int m)//check if you can place number - if you can, return number of overlaps
{
int c=1;
while(v)//extract digits one by one
{
if(vnum[x][y] && (v%10)!=val[x][y])
return 0;
if(vnum[x][y])
c++;
v/=10;
x+=mx[m];
y+=my[m];
}
return c;
}
void apply(int x,int y,int v,int m)//place number - no sanity checks
{
while(v)//extract digits one by one
{
val[x][y]=v%10;
vnum[x][y]++;
v/=10;
x+=mx[m];
y+=my[m];
}
}
void deapply(int x,int y,int v,int m)//remove number - no sanity checks
{
while(v)
{
vnum[x][y]--;
v/=10;
x+=mx[m];
y+=my[m];
}
}
int best=100;
void recur(int num)//go down a semi-random path
{
if(num<best)
{
best=num;
if(best)
printf("FAILED AT %d\n",best);
else
printf("SUCCESS\n");
for(int x=5;x<16;x++) // 16 and 16
{
for(int y=5;y<16;y++)
{
if(vnum[x][y]==0)
putchar('.');
else
putchar(val[x][y]+'0');
}
putchar('\n');
}
fflush(stdout);
}
if(num==0)
return;
int s=num*num,t;
vector<int> poss;
for(int x=5;x<16;x++)
for(int y=5;y<16;y++)
for(int m=0;m<4;m++)
if(t=check(x,y,s,m))
poss.push_back((x)|(y<<8)|(m<<16)|(t<<24));//compress four numbers into an int
if(poss.size()==0)
return;
sort(poss.begin(),poss.end());//essentially sorting by t
t=poss.size()-1;
while(t>=0 && (poss[t]>>24)==(poss.back()>>24))
t--;
t++;
//t is now equal to the smallest index which has the maximal overlap
t=poss[rand()%(poss.size()-t)+t];//select random index>=t
apply(t%256,(t>>8)%256,s,(t>>16)%256);//extract random number
recur(num-1);//continue down path
}
int main()
{
srand((unsigned)time(0));//seed
while(true)
{
for(int i=0;i<21;i++)//reset board
{
memset(val[i],-1,21*sizeof(int));
memset(vnum[i],-1,21*sizeof(int));
}
for(int i=5;i<16;i++)
{
memset(val[i]+5,0,11*sizeof(int));
memset(vnum[i]+5,0,11*sizeof(int));
}
recur(100);
}
}
Using a random search so far I only got to 92 squares with one unused spot (8 missing numbers: 5041 9025 289 10000 4356 8464 3364 3249)
1 5 2 1 2 9 7 5 6 9 5
6 1 0 8 9 3 8 4 4 1 2
9 7 2 2 5 0 0 4 8 8 2
1 6 5 9 6 0 4 4 7 7 4
4 4 2 7 6 1 2 9 0 2 2
2 9 6 1 7 8 4 4 0 9 3
6 5 5 3 2 6 0 1 4 0 6
4 7 6 1 8 1 1 8 2 8 1
8 0 1 3 4 8 1 5 3 2 9
0 5 9 6 9 8 8 6 7 4 5
6 6 2 9 1 7 3 9 6 9
The algorithm basically uses as solution encoding a permutation on the input (search space is 100!) and then places each number in the "topmost" legal position. The solution value is measured as the sum of the squares of the lengths of the numbers placed (to give more importance to long numbers) and the number of "holes" remaining (IMO increasing the number of holes should raise the likehood that another number will fit in).
The code has not been optimized at all and is only able to decode a few hundred solutions per second. Current solution has been found after 196k attempts.
UPDATE
Current best solution with this approach is 93 without free holes (7 missing numbers: 676 7225 3481 10000 3364 7744 5776):
9 6 0 4 8 1 0 0 9 3 6
6 4 0 0 2 2 5 6 8 8 9
1 7 2 9 4 1 5 4 7 6 3
5 8 2 3 8 6 4 9 6 5 7
2 4 4 4 1 8 2 8 2 7 2
1 0 8 9 9 1 3 4 4 9 1
2 1 2 9 6 1 0 6 2 4 1
2 3 5 5 3 9 9 4 0 9 6
5 0 0 6 1 0 3 5 2 0 3
2 7 0 4 2 2 5 2 8 0 9
9 8 2 2 6 5 3 4 7 6 1
This is a solution (all 100 numbers placed) however using a 12x12 grid (MUCH easier)
9 4 6 8 7 7 4 4 5 5 1 7
8 3 0 5 5 9 2 9 6 7 6 4
4 4 8 3 6 2 6 0 1 7 8 4
4 8 4 2 9 1 4 0 5 6 1 4
9 1 6 9 4 8 1 5 4 2 0 1
9 4 4 7 2 2 5 2 2 5 0 0
4 6 2 2 5 8 4 2 7 4 0 2
0 3 3 3 6 4 0 0 6 3 0 9
9 8 0 1 2 1 7 9 5 5 9 1
6 8 4 2 3 5 2 6 3 2 0 6
9 9 8 2 5 2 9 9 4 2 2 7
1 1 5 6 6 1 9 3 6 1 5 4
It has been found using a truly "brute force" approach, starting from a random matrix and keeping randomly changing digits when that improved the coverage.
This solution has been found by an highly unrolled C++ program automatically generated by a Python script.
Update 2
Using an incremental approach (i.e. keeping a more complex data structure so that when changing a matrix element the number of targets covered can be updated instead than recomputed) I got a much faster search (about 15k matrices/second investigated with a Python implementation running with PyPy).
In a few minutes this version was able to find a 99 quasi-solution (a number is still missing):
7 0 5 6 5 1 1 5 7 1 6
4 6 3 3 9 8 8 6 7 6 1
3 9 0 8 2 6 1 1 4 7 8
1 1 0 8 9 9 0 0 4 4 6
3 4 9 0 4 9 0 4 6 7 1
6 4 4 6 8 6 3 2 5 2 9
9 7 8 4 1 1 4 0 5 4 2
6 2 4 1 5 2 2 1 2 9 7
9 8 2 5 2 2 7 3 6 5 0
3 1 2 5 0 0 6 3 0 5 4
7 5 6 9 2 1 6 5 3 4 6
UPDATE 3
Ok. After a some time (no idea how much) the same Python program actually found a complete solution (several ones indeed)... here is one
6 4 6 9 4 1 2 9 7 3 6
9 2 7 7 4 4 8 1 2 1 7
1 0 6 2 7 0 4 4 8 3 4
2 1 2 2 5 5 9 2 9 6 5
9 2 5 5 2 0 2 6 3 9 1
1 6 3 6 0 0 9 3 7 0 6
6 0 0 4 9 0 1 6 0 0 4
9 8 4 4 8 0 1 4 5 2 3
2 4 8 2 8 1 6 8 6 7 5
1 7 6 9 2 4 5 4 2 7 6
6 6 3 8 8 5 6 1 5 2 1
The searching program can be found here...
You've got 100 numbers and 121 cells to work with, so you'll need to be very efficient. We should try to build up the grid, so that each time we fill a cell, we attain a new number in our list.
For now, let's only worry about 68 4-digit numbers. I think a good chunk of the shorter numbers will be in our grid without any effort.
Start with a 3x3 or 4x4 set of numbers in the top-left of your grid. It can be arbitrary, or fine-tune for slightly better results. Now let's fill in the rest of the grid one square at a time.
Repeat these steps:
Fill an empty cell with a digit
Check which numbers that knocked off the list
If it didn't knock off any 4-digit numbers, try a different digit or cell
Eventually you may need to fill 2 cells or even 3 cells to achieve a new 4-digit number, but this should be uncommon, except at the end (at which point, hopefully there's a lot of empty space). Continue the process for the (few?) remaining 3-digit numbers.
There's a lot room for optimizations and tweaks, but I think this technique is fast and promising and a good starting point. If you get an answer, share it with us! :)
Update
I tried my approach and only got 87 out of the 100:
10894688943
60213136008
56252211674
61444925224
59409675697
02180334817
73260193640
.5476685202
0052034645.
...4.948156
......4671.
My guess is that both algorithms are too slow. Some optimization algorithm might work like best-first search or simulated annealing but my I don't have much experience on programming those.
Have you tried any primary research on Two-Dimensional Bin Packing (2DBP) algorithms? Google Scholars is a good start. I did this a while ago when building an application to generate mosaics.
All rectangular bin packing algorithms can be divided into 4 groups based on their support for the following constraints:
Must the resulting bin be guillotine cuttable? I.e. do you have to later slice the bin(s) in half until all the pieces are unpacked?
Can the pieces be rotated to fit into the bin? Not an issue with square pieces, so this makes more algorithms available to you.
Out of all the algorithms I looked into, the most efficient solution is an Alternate Directions (AD) algorithm with a Tabu Search optimization layer. There are dissertations which prove this. I may be able to dig-up some links if this helps.
Some ideas off the top of my head, without investing much time into thinking about details.
I would start by counting the number of occurrences of each digit in all squares 1..100. The total number of digits will be obviously larger than 121, but by analyzing individual frequencies you can deduce which digits must be grouped on a single line to form as many different squares as possible. For example, if 0 has the highest frequency, you have to try to put as many squares containing a 0 on the same line.
You could maintain a count of digits for each line, and each time you place a digit, you update the count. This lets you easily compute which square numbers have been covered by that particular line.
So, the program will still be brute-force, but it will exploit the problem structure much better.
PS: Counting digit frequencies is the easiest way to decide whether a certain permutation of digits constitutes a square.