Optimizing loop by using floating point value as loop counter - loops

I need to execute a loop
while (X) do Y
for a N times, which is a large number. While Y, the loop body, is rather quick, the test X takes up about 70% of the runtime.
I can calculate the number of loop iterations N beforehand so instead of using X as the condition, a simple For-Loop would be possible.
for (i=1 to N) do Y
However, N might exceed the maximum value an integer can store on the machine, so this is not an option. As an alternative, I proposed to use a floating point variable F instead. As N is large, it is likely that F cannot be exactly equal to N. Thus, I calculate F to be the largest floating point number smaller than N. This allows me to run
for (i=1 to F) do Y
while (X) do Y
Most of the iterations will not need to test X everytime, only the last N-F do.
The question is: How would I implement a for-Loop from 1 to F? Simply increasing a counter or decreasing F by 1 in each step would not work, as the numerical error would grow too large. My current solution is:
for (while F > MAXINT)
for (i=1 to MAXINT)
do Y
F -= MAXINT
while (X) do Y
Is there a better way to solve this problem?

What do you mean by numerical error? Floating point counting is exact within its precision.
Here are the maximum values representable by integers exactly using the following data types:
uint32max = 4294967295
uint64max = 18446744073709551615
float32intmax = 16777216
float64intmax = 9007199254740992
Every integer from 0 to the max is exactly representable without numerical error.
As you can see, the largest count is available with uint64. Next comes float64, then uint32 and finally float32.
What happens when you increment uint32=4294967295 ? 0.
What happens when you increment float32=16777216 ? 16777216.
Which is the better behavior?
Have you thought about using a 2-dimensional loop? If N ism the maximum count for a 1-dimensional loop, then N x N is the maximum for a 2-dimensional loop, etc. so that if you maximum count is less than MAXUINT x MAXUINT, then decompose you number N such that:
N == M * MAXUINT + R;
where
M = N / MAXUINT;
R = N - M * MAXUINT;
then
for (i = M; i--;) for (j = MAXUINT; j--;) DoStuff();
for (i = R; i--;) DoStuff();
If MAXUINT*MAXUINT is not a large enough count for you, you can add 3-, 4-, ... -dimensional loops.

Related

Find Minimum Operand to Maximize sum of bitwise AND operator

Given an array of integers Arr and an integer K, bitwise AND is to be performed on each element A[i] with an integer X
Let Final sum be defined as follows:
Sum of ( A[i] AND X ) for all values of i ( 0 to length of array-1 )
Return the integer X subject to following constraints:
Final sum should be maximum
X should contain exactly K bits as 1 in its binary representation
If multiple values of X satisfy the above conditions, return the minimum possible X
Input:
Arr : [8,4,2]
K = 2
Output: X=12
12 Contains exactly 2 bits in its binary and is the smallest number that gives maximum possible answer for summation of all (A[i] AND X)
Approach Tried :
Took bitwise OR for all numbers in the array in binary and retained the first K bits of the binary that had 1 , made remaining bits 0, convert back to int
Passed 7/12 Test Cases
Can someone help me out with what mistake am I making with regards to the approach or suggest a better approach ? Thanks in advance.
Consider an input like [ 8, 4, 4, 4 ], K = 1. Your algorithm will give 8 but the correct answer is 4. Just because a given bit is more significant doesn't mean that it will automatically contribute more to the sum, as there might be more than twice as many elements of the array that use a smaller bit.
My suggestion would be to compute a weight for each bit of your potential X -- the number of elements of the array that have that bit set times the value of that bit (2i for bit i). Then find the K bits with the largest weight.
To do this, you need to know how big your integers are -- if they are 32 bits, you need to compute just 32 weights. If they might be bigger you need more. Depending on your programming language you may also need to worry about overflow with your weight calculations (or with the sum calculation -- is this a true sum, or a sum mod 2n for some n?). If some elements of the array might be negative, how are negatives represented (2s complement?) and how does that interact with AND?
Let dp[k][i] represent the maximum sum(a & X), a ∈ array, where i is the highest bit index in X and k is the number of bits in X. Then:
dp[1][i]:
sum(a & 2^i)
dp[k][i]:
sum(a & 2^i) + max(dp[k-1][j])
for j < i
sum(a & 2^i) can be precalculated for all values of i in O(n * m), where m is the word size. max(dp[k-1][j]) is monotonically increasing over j and we want to store the earliest instance of each max to minimise the resulting X.
For each k, therefore, we iterate over m is. Overall time complexity O(k * m + n * m), where m is the word size.

Maximum possible integer x, such that sum of every element of array divided by x is not less than k

We are given an array of n integers, and a constant value k, can any one
suggest me to find out the maximum possible integer x such that arr[0]/x + arr[1]/x +.. arr[n-1]/x >=k ,
-> where '/' is the integer division
-> sum of all elements of array >= k
-> k is a constant(1<=k<=10^5)
-> 1<=n<=10^5.
e.g. n=5, k=3
arr=[1,1,1,8,8]
answer-> x=4
in something like o(N log N) ?
Here is an algorithm that often meets your bound on time efficiency. I assume that your array values are non-negative. The algorithm depends on these facts:
Your objective function arr[0]/x + arr[1]/x +.. arr[n-1]/x (let's call it f(x)) is a decreasing function of x. In other words, if x increases then f(x) will stay the same or decrease.
f(1) equals the sum of the elements of the array, so f(1) >= k. In other words, at x = 1 the objective function is not below the target value k.
If M is set to the maximum array value, the value of arr[i] // (M + 1) is zero, so f(M + 1) = 0. In other words, at x = M + 1 the objective is below the target value k.
So we have upper and lower bounds on the value of x for a decreasing function. We can therefore do a binary search from 1 to M + 1 for the value of x where
f(x) >= k and f(x + 1) < k
Only one value of x will satisfy that, and a binary search can easily find it. The binary search will take log(M) steps. Each step involves one evaluation of f(x) which takes N steps to use each array member. Thus the overall time efficiency is O(N log(M)). If M (the maximum array value) is of the order of N then that is your desired efficiency. At the limiting values you give for N and the array values, we have M < N^2, so N log(M) < 2 N log(N) and your desired efficiency is still met. If N is small and M is large, your desired efficiency is not met. (This means an array like [10^9, 10^9-1] where N = 2 and M = 10^9 which could take 30 steps in the binary search.) This may or may not meet your needs.

While loop time complexity O(logn)

I cannot understand why the time complexity for this code is O(logn):
double n;
/* ... */
while (n>1) {
n*=0.999;
}
At least it says so in my study materials.
Imagine if the code were as follows:
double n;
/* ... */
while (n>1) {
n*=0.5;
}
It should be intuitive to see how this is O(logn), I hope.
When you multiply by 0.999 instead, it becomes slower by a constant factor, but of course the complexity is still written as O(logn)
You want to calculate how many iterations you need before n becomes equal to (or less than) 1.
If you call the number of iterations for k you want to solve
n * 0.999^k = 1
It goes like this
n * 0.999^k = 1
0.999^k = 1/n
log(0.999^k) = log(1/n)
k * log(0.999) = -log(n)
k = -log(n)/log(0.999)
k = (-1/log(0.999)) * log(n)
For big-O we only care about "the big picture" so we throw away constants. Here log(0.999) is a negative constant so (-1/log(0.999)) is a positive constant that we can "throw away", i.e. set to 1. So we get:
k ~ log(n)
So the code is O(logn)
From this you can also notice that the value of the constant (i.e. 0.999 in your example) doesn't matter for the big-O calculation. All constant values greater than 0 and less than 1 will result in O(logn).
Logarithm has two inputs: a base and a number. The result of a logarithm is the power you need to raise the base to to achieve the given number. Since your base is 0.999, the number is the first smaller than 1 and you have a scalar, which is n, effectively the number of steps depends on the power you need to raise your base to achieve such a small number, which multiplied with n will yield a smaller number than 1. This corresponds to the definition of the logarithm, with which I have started my answer.
EDIT:
Think about it this way: You have n as an input and you search for the first k where
n * .999^k < 1
since you are searching k by incrementing it, since if you have l >= n at a step, then you will have l * .999 in the next step. Repeating this achieves a logarithmic complexity for your multiplication algorithm.

Integer Division Algorithm Analysis

For an assignment we are required to write a division algorithm in order to complete a certain question using just addition and recursion. I found that, without using tail recursion, the naive repeated subtraction implementation can easily result in a stack overflow. So doing a quick analysis of this method, and correct me if I'm wrong, shows that if you divide A by B, with n and m binary digits respectively, it should be exponential in n-m. I actually get
O( (n-m)*2^(n-m) )
since you need to subtract an m binary digit number from an n binary digit number 2^(n-m) times in order to drop the n digit number to an n-1 digit number, and you need to do this n-m times to get a number with at most m digits in the repeated subtraction division, so the runtime should be as mentioned. Again, I very well may be wrong so someone please correct me if I am. This is assuming O(1) addition since I'm working with fixed size integers. I suppose with fixed size integers one could argue the algorithm is O(1).
Back to my main question. I developed a different method to perform integer division which works much better, even when using it recursively, based on the idea that for
P = 2^(k_i) + ... 2^(K_0)
we have
A/B = (A - B*P)/B + P
The algorithm goes as follows to caclulate A/B:
input:
A, B
i) Set Q = 0
ii) Find the largest K such that B * 2^K <= A < B * 2(K + 1)
iii) Q -> Q + 2^K
iv) A -> A - B * 2^k
v) Repeat steps ii) through iv) until A <= B
vi) Return Q (and A if you want the remainder)
with the restrictions of using only addition, I simply add B to itself on each recursive call, however here is my code without recursion and with the use of shifts instead of addition.
int div( unsigned int m, unsigned int n )
{
// q is a temporary n, sum is the quotient
unsigned int q, sum = 0;
int i;
while( m > n )
{
i = 0;
q = n;
// double q until it's larger than m and record the exponent
while( q <= m )
{
q <<= 1;
++i;
}
i--;
q >>= 1; // q is one factor of 2 too large
sum += (1<<i); // add one bit of the quotient
m -= q; // new numerator
}
return sum;
}
I feel that sum |= (1<<i) may be more appropriate in order to emphasize I'm dealing with a binary representation, but it didn't seem to give any performance boost and may make it harder to understand. So, if M and N are the number of bits in m and n respectively, an analysis suggests the inner loop is performed M - N times and each time the outer loop is completed that m looses one bit, and it must also be completed M - N times in order for the condition m <= n so I get that it's O( (M - N)^2 ).
So after all of that, I am asking if I am correct about the runtime of the algorithm and whether it can be improved upon?
Your algorithm is pretty good and your analysis of the running time is correct, but you don't need to do the inner loop every time:
unsigned div(unsigned num, unsigned den)
{
//TODO check for divide by zero
unsigned place=1;
unsigned ret=0;
while((num>>1) >= den) //overflow-safe check
{
place<<=1;
den<<=1;
}
for( ;place>0; place>>=1,den>>=1)
{
if (num>=den)
{
num-=den;
ret+=place;
}
}
return ret;
}
That makes it O(M-N)

A numbers power between 0 and 1 in C

I'm making a program to replace math.h's pow() function.
I'm not using any functions from math.h.
The problem is, I can calculate powers as integers like
15-2
45.3211
but I can't calculate
x2.132
My program first finds integer power of x (x2) and multiplies it by (x0.132).
I know that x0.132 is 1000th root of x to the power 132 but I can't solve it.
How can I find xy (0 < y < 1)
To compute x ^ y, 0 < y < 1 :
Approximate y as a rational fraction, (a/b)
(Easiest way: Pick whatever b you want to get sufficient accuracy as a constant.
Then use: a = b * y.)
Approximate the b root of y using any method you like, such as Newton's.
(Simplest way: You know it's between 0 and b and can easily tell if a given value is too low or too high. So keep a min that starts at zero and a max that starts at b. Repeatedly try (min + max) / 2, see if it's too big or too small, and adjust min or max appropriately. Repeat until min and max are nearly the same.)
Raise that to the a power.
(Possibly by repeatedly multiplying it by itself. Optimize this if you like. For example, a^4 can be computed with just two multiplications, one to find a^2 and then one to square it. This generalizes easily.)
Use the factorization inherent in floating point formats to split x=2^e*m with 1<=m<2 to create the sub-problems 2^(e*y) and m^y
Use square roots, x^y=sqrt(x)^(2*y) and if there is an integer part in 2*b, split that off.
Use the binomial theorem for x close to 1, which will occur when iterating the square root.
(1+h)^y=1+y*h+(y*(y-1))/2*h^2+...+binom(y,j)*h^j+...
where the quotient from one term to the next is (y-j)/(j+1)*h
h=x-1;
term = y*h;
sum = 1+term;
j=1;
while(1+term !=1) {
term *= h*(y-j)/(1+j);
sum += term;
j+=1;
}

Resources