Algorithm to break up a vector

[No Such Luck]

Hi All:

I'm not sure if this is the right place to ask this question, but I
couldn't find a more appropriate group. This is more of a theory
question regarding an algorithm implemented in C, not necessarily a C
language question.

I'm trying to break up a vector into an arbitrary number of subvectors,
equal (or as near to equal) in size as possible. My problem occurs when
the vector is not evenly divisible by the number of subvectors (i.e.,
vector length: 45, number of subvectors: 7)

If I break up the vector based on (int)(vectorsiz e/numsubvectors), not
all of the original vector data will be included. If I break up the
vector based on ceil(vectorsize/numsubvectors), the end of the vector
will be passed, resulting in a memory violation.

The obvious solution to the above example be to have 3 subvectors of
size 7, and 4 subvectors of size 6, but I'm having a hard time thinking
of a way to implement this for an arbitrary vector size and number of
subvectors. I know the mod(%) operator is probably required, and I know
for the above example that the algorithm I need will result:

Subvector 1: Size 6
Subvector 2: Size 7
Subvector 3: Size 6
Subvector 4: Size 7
Subvector 5: Size 6
Subvector 6: Size 7
Subvector 7: Size 6

Anyone have any ideas?

Thanks.

[Michael Mair]

No Such Luck wrote:

Hi All:

I'm not sure if this is the right place to ask this question, but I
couldn't find a more appropriate group. This is more of a theory
question regarding an algorithm implemented in C, not necessarily a C
language question.
If you give us C code to work with, then we will point out
possible algorithmic shortcomings.
A request for a C algorithm is not topical here.
comp.programmin g might be a good starting point to ask for
an algorithm. Then you can implement it -- and if you have
problems because your program does not run as intended,
then you can isolate a minimal running program exhibiting
these problems and we will help you.

I'm trying to break up a vector into an arbitrary number of subvectors,
equal (or as near to equal) in size as possible. My problem occurs when
the vector is not evenly divisible by the number of subvectors (i.e.,
vector length: 45, number of subvectors: 7)

If I break up the vector based on (int)(vectorsiz e/numsubvectors), not
all of the original vector data will be included. If I break up the
vector based on ceil(vectorsize/numsubvectors), the end of the vector
will be passed, resulting in a memory violation.

The obvious solution to the above example be to have 3 subvectors of
size 7, and 4 subvectors of size 6, but I'm having a hard time thinking
of a way to implement this for an arbitrary vector size and number of
subvectors. I know the mod(%) operator is probably required, and I know
for the above example that the algorithm I need will result:

Subvector 1: Size 6
Subvector 2: Size 7
Subvector 3: Size 6
Subvector 4: Size 7
Subvector 5: Size 6
Subvector 6: Size 7
Subvector 7: Size 6

Anyone have any ideas?

Yes. This sounds like an idiotic homework question. More natural
would be to give the maximum length of subvectors and to go for
7|7|7|6|6|6|6 instead of 6|7|6|7|6|7|6.

Hints:
45/7 = 6
45%7 = 3
Cheers
Michael
--
E-Mail: Mine is a gmx dot de address.

[Eric Sosman]

No Such Luck wrote:

Hi All:

I'm not sure if this is the right place to ask this question, but I
couldn't find a more appropriate group. This is more of a theory
question regarding an algorithm implemented in C, not necessarily a C
language question.

I'm trying to break up a vector into an arbitrary number of subvectors,
equal (or as near to equal) in size as possible. My problem occurs when
the vector is not evenly divisible by the number of subvectors (i.e.,
vector length: 45, number of subvectors: 7)

If I break up the vector based on (int)(vectorsiz e/numsubvectors), not
all of the original vector data will be included. If I break up the
vector based on ceil(vectorsize/numsubvectors), the end of the vector
will be passed, resulting in a memory violation.

The obvious solution to the above example be to have 3 subvectors of
size 7, and 4 subvectors of size 6, but I'm having a hard time thinking
of a way to implement this for an arbitrary vector size and number of
subvectors. I know the mod(%) operator is probably required, and I know
for the above example that the algorithm I need will result:

Subvector 1: Size 6
Subvector 2: Size 7
Subvector 3: Size 6
Subvector 4: Size 7
Subvector 5: Size 6
Subvector 6: Size 7
Subvector 7: Size 6

An easy way to proceed is (pseudocode):

int vecsize = length of original vector;
int subcount = number of sub-vectors desired;
int subsizes[subcount];
for ( ; subcount > 0; --subcount) {
subsizes[subcount-1] =
(2 * vecsize + subcount) / (2 * subcount);
vecsize -= subsizes[subcount-1];
}

(That mess in the middle is merely "vecsize / subcount,
rounded." You could use an ordinary integer division or
a "ceiling" without affecting the validity of the method,
but you'd get a different pattern of long and short
sub-vectors.)

This method will balance the sub-vector lengths as
evenly as is possible. Its principal virtue is that it's
easy to see why it works, and hence hard to get wrong.

<off-topic>

It's also an illustrative example of a divide-and-conquer
solution that *doesn't* beg for a recursive implementation.

</off-topic>

--
Er*********@sun .com

[No Such Luck]

"Eric Sosman" <er*********@su n.com> wrote in message
news:cp******** **@news1brm.Cen tral.Sun.COM...

No Such Luck wrote:

Hi All:

I'm not sure if this is the right place to ask this question, but I
couldn't find a more appropriate group. This is more of a theory
question regarding an algorithm implemented in C, not necessarily a C
language question.

I'm trying to break up a vector into an arbitrary number of subvectors,
equal (or as near to equal) in size as possible. My problem occurs when
the vector is not evenly divisible by the number of subvectors (i.e.,
vector length: 45, number of subvectors: 7)

If I break up the vector based on (int)(vectorsiz e/numsubvectors), not
all of the original vector data will be included. If I break up the
vector based on ceil(vectorsize/numsubvectors), the end of the vector
will be passed, resulting in a memory violation.

The obvious solution to the above example be to have 3 subvectors of
size 7, and 4 subvectors of size 6, but I'm having a hard time thinking
of a way to implement this for an arbitrary vector size and number of
subvectors. I know the mod(%) operator is probably required, and I know
for the above example that the algorithm I need will result:

Subvector 1: Size 6
Subvector 2: Size 7
Subvector 3: Size 6
Subvector 4: Size 7
Subvector 5: Size 6
Subvector 6: Size 7
Subvector 7: Size 6

An easy way to proceed is (pseudocode):

int vecsize = length of original vector;
int subcount = number of sub-vectors desired;
int subsizes[subcount];
for ( ; subcount > 0; --subcount) {
subsizes[subcount-1] =
(2 * vecsize + subcount) / (2 * subcount);
vecsize -= subsizes[subcount-1];
}

(That mess in the middle is merely "vecsize / subcount,
rounded." You could use an ordinary integer division or
a "ceiling" without affecting the validity of the method,
but you'd get a different pattern of long and short
sub-vectors.)

This method will balance the sub-vector lengths as
evenly as is possible. Its principal virtue is that it's
easy to see why it works, and hence hard to get wrong.

<off-topic>

It's also an illustrative example of a divide-and-conquer
solution that *doesn't* beg for a recursive implementation.

</off-topic>

Thanks, Eric. I will give this implementation a whirl tomorrow. A divide and
conquer technique never occurred to me. I thought I would have to determine
the size of all the subvectors in the outset...

[Mike Wahler]

"No Such Luck" <[e-mail] > wrote in message
news:BZ******** ************@rc n.net...

Thanks, Eric. I will give this implementation a whirl tomorrow. A divide and conquer technique never occurred to me.

It should have. :-)

'Divide and conquer' is one of the cornerstones of programming,
in any language.

-Mike

[Lawrence Kirby]

On Thu, 09 Dec 2004 21:22:08 -0500, No Such Luck wrote:

....

The obvious solution to the above example be to have 3 subvectors of
size 7, and 4 subvectors of size 6, but I'm having a hard time thinking
of a way to implement this for an arbitrary vector size and number of
subvectors. I know the mod(%) operator is probably required, and I know
for the above example that the algorithm I need will result:
You can do this using modulo arithmetic and I give some example code
below. It doesn't use % to perform the modulo arithmetic but it could do.
Subvector 1: Size 6
Subvector 2: Size 7
Subvector 3: Size 6
Subvector 4: Size 7
Subvector 5: Size 6
Subvector 6: Size 7
Subvector 7: Size 6

An easy way to proceed is (pseudocode):

int vecsize = length of original vector; int subcount = number of
sub-vectors desired; int subsizes[subcount]; for ( ; subcount > 0;
--subcount) {
subsizes[subcount-1] =
(2 * vecsize + subcount) / (2 * subcount);
vecsize -= subsizes[subcount-1];
}
}
(That mess in the middle is merely "vecsize / subcount, rounded." You
could use an ordinary integer division or a "ceiling" without affecting
the validity of the method, but you'd get a different pattern of long
and short sub-vectors.)

This method will balance the sub-vector lengths as
evenly as is possible.
It doesn't distribute them evenly. For example with vecsize=45 and
subcount=14 I get subvectors from it like this:

3 4 3 4 3 4 3 3 3 3 3 3 3 3
Its principal virtue is that it's easy to see
why it works, and hence hard to get wrong.
I can see that it probably works, but I haven't proved it to my
satisfaction yet. This isn't as simple as it first looks as the uneven
distribution indicates. Essentially it creates a non-linear
convergence.
<off-topic>

It's also an illustrative example of a divide-and-conquer
solution that *doesn't* beg for a recursive implementation.
If this qualifies as a divide and conquer algorithm it is a degenerate
one, more a lop one off the end and conquer. :-) You might just as well
say that a linear search (as well as many simple loops) is divide and
conquer because you can test the first (or last) element and then if
necessary do a linear search on the rest. Compare that to a binary search.
</off-topic>

Thanks, Eric. I will give this implementation a whirl tomorrow. A divide
and conquer technique never occurred to me. I thought I would have to
determine the size of all the subvectors in the outset...

Try this. It does produce an even (or as even as possible) distribution of
subvector sizes, although a small amount of bias can be controlled by the
initial value of sum. It doesn't write the results to an array but you can
do so easily if you want. With vecsize=45 and subcount=14 it produces

3 3 4 3 3 3 4 3 3 3 3 4 3 3
static void subsizes(int vecsize, int subcount)
{
int basesize = vecsize / subcount;
int bumps = vecsize % subcount;
int sum = subcount/2; /* Try also 0, subcount-1, (subcount+1)/2 */
int testtotal = 0; /* For checking only */
int i;

for (i = 0; i < subcount; i++) {
int subsize = basesize;

if ((sum += bumps) >= subcount) {
sum -= subcount;
subsize++;
}

printf(" %d", subsize);
testtotal += subsize;
}

printf("\n\ntes ttotal=%d\n", testtotal);
}

Lawrence

[No Such Luck]

Lawrence Kirby wrote:

On Thu, 09 Dec 2004 21:22:08 -0500, No Such Luck wrote:

...

The obvious solution to the above example be to have 3 subvectors of size 7, and 4 subvectors of size 6, but I'm having a hard time thinking of a way to implement this for an arbitrary vector size and number of subvectors. I know the mod(%) operator is probably required, and I know for the above example that the algorithm I need will result:
You can do this using modulo arithmetic and I give some example code
below. It doesn't use % to perform the modulo arithmetic but it could do. Subvector 1: Size 6
Subvector 2: Size 7
Subvector 3: Size 6
Subvector 4: Size 7
Subvector 5: Size 6
Subvector 6: Size 7
Subvector 7: Size 6

An easy way to proceed is (pseudocode):

int vecsize = length of original vector; int subcount = number of
sub-vectors desired; int subsizes[subcount]; for ( ; subcount > 0; --subcount) {
subsizes[subcount-1] =
(2 * vecsize + subcount) / (2 * subcount);
vecsize -= subsizes[subcount-1];
}
}
(That mess in the middle is merely "vecsize / subcount, rounded." You could use an ordinary integer division or a "ceiling" without affecting the validity of the method, but you'd get a different pattern of long and short sub-vectors.)

This method will balance the sub-vector lengths as
evenly as is possible.
It doesn't distribute them evenly. For example with vecsize=45 and
subcount=14 I get subvectors from it like this:

3 4 3 4 3 4 3 3 3 3 3 3 3 3
Its principal virtue is that it's easy to see
why it works, and hence hard to get wrong.

I can see that it probably works, but I haven't proved it to my
satisfaction yet. This isn't as simple as it first looks as the

uneven distribution indicates. Essentially it creates a non-linear
convergence.

It works perfectly. I have implemented it and tried in on thousands of
variations of vecsize and subcount, all with correct results. And it
doesn't matter that the distribution of subsizes is uneven. The only
stipulation is that the sizes of all the subvectors are either n or
n+1, and that their sum is the original vecsize.

[Lawrence Kirby]

On Wed, 15 Dec 2004 13:20:21 -0800, No Such Luck wrote:

....

I can see that it probably works, but I haven't proved it to my
satisfaction yet. This isn't as simple as it first looks as the uneven

distribution indicates. Essentially it creates a non-linear
convergence.

It works perfectly. I have implemented it and tried in on thousands of
variations of vecsize and subcount, all with correct results. And it

I'm not saying it doesn't work, I'm pretty sure that it does. My question
is whether you can PROVE that it is correct. What you have done is good
verification but unless you have tested EVERY possible variation it isn't
proof.
doesn't matter that the distribution of subsizes is uneven. The only
stipulation is that the sizes of all the subvectors are either n or
n+1, and that their sum is the original vecsize.

In which case you can solve this very simply, for example

static void subsizes2(int vecsize, int subcount)
{
int basesize = vecsize / subcount;
int bumps = vecsize % subcount;
int i;

for (i = 0; i < subcount; i++) {
printf(" %d", basesize + (i < bumps));
}

putchar('\n');
}

[No Such Luck]

Lawrence Kirby wrote:

On Wed, 15 Dec 2004 13:20:21 -0800, No Such Luck wrote:

...

I can see that it probably works, but I haven't proved it to my
satisfaction yet. This isn't as simple as it first looks as the uneven

distribution indicates. Essentially it creates a non-linear
convergence.

It works perfectly. I have implemented it and tried in on thousands of variations of vecsize and subcount, all with correct results. And

it
I'm not saying it doesn't work, I'm pretty sure that it does. My question is whether you can PROVE that it is correct. What you have done is good verification but unless you have tested EVERY possible variation it isn't proof.

Like I said, I've tested Eric's algorithm on hundreds of vecsizes
ranging from 100 to 1500, and subcounts ranging from 1 to 100 (i.e.,
thousands of trials). All with correct results (all verified
numerically, and some verified visually).

What proof are you looking for? Something like:

for (vecsize = 1; vecsize < 1000000; vecsize++)
{
for (subcount = 1; soubcount < 1000000; subcount++)
{
test_algorithm( vecsize, subcount);
}
}

[Mark McIntyre]

On 16 Dec 2004 10:49:33 -0800, in comp.lang.c , "No Such Luck"
<no*********@ho tmail.com> wrote:

Like I said, I've tested Eric's algorithm on hundreds of vecsizes
ranging from 100 to 1500, and subcounts ranging from 1 to 100 (i.e.,
thousands of trials). All with correct results (all verified
numerically, and some verified visually).
This isn't proof tho, its empirical evidence. A similar huge body of
evidence exists to support Newton's laws. They're wrong.
What proof are you looking for?
You seem to be trying to prove an algo. I'd suggest a mathematical method.
Something like:

for (vecsize = 1; vecsize < 1000000; vecsize++)
{
for (subcount = 1; soubcount < 1000000; subcount++)
{
test_algorithm( vecsize, subcount);
}
}

Thats still empirical.
--
Mark McIntyre
CLC FAQ <http://www.eskimo.com/~scs/C-faq/top.html>
CLC readme: <http://www.ungerhu.com/jxh/clc.welcome.txt >

----== Posted via Newsfeeds.Com - Unlimited-Uncensored-Secure Usenet News==----
http://www.newsfeeds.com The #1 Newsgroup Service in the World! 120,000+ Newsgroups
----= East and West-Coast Server Farms - Total Privacy via Encryption =----