Related
Having some difficulty troubleshooting code I wrote in C to perform a logistic regression. While it seems to work on smaller, semi-randomized datasets, it stops working (e.g. assigning proper probabilities of belonging to class 1) at around the point where I pass 43,500 observations (determined by tweaking the number of observations created. When creating the 150 features used in the code, I do create the first two as a function of the number of observations, so I'm not sure if maybe that's the issue here, though I am using double precision. Maybe there's an overflow somewhere in the code?
The below code should be self-contained; it generates m=50,000 observations with n=150 features. Setting m below 43,500 should return "Percent class 1: 0.250000", setting to 44,000 or above will return "Percent class 1: 0.000000", regardless of what max_iter (number of times we sample m observations) is set to.
The first feature is set to 1.0 divided by the total number of observations, if class 0 (first 75% of observations), or the index of the observation divided by the total number of observations otherwise.
The second feature is just index divided by total number of observations.
All other features are random.
The logistic regression is intended to use stochastic gradient descent, randomly selecting an observation index, computing the gradient of the loss with the predicted y using current weights, and updating weights with the gradient and learning rate (eta).
Using the same initialization with Python and NumPy, I still get the proper results, even above 50,000 observations.
#include <stdlib.h>
#include <stdio.h>
#include <math.h>
#include <time.h>
// Compute z = w * x + b
double dlc( int n, double *X, double *coef, double intercept )
{
double y_pred = intercept;
for (int i = 0; i < n; i++)
{
y_pred += X[i] * coef[i];
}
return y_pred;
}
// Compute y_hat = 1 / (1 + e^(-z))
double sigmoid( int n, double alpha, double *X, double *coef, double beta, double intercept )
{
double y_pred;
y_pred = dlc(n, X, coef, intercept);
y_pred = 1.0 / (1.0 + exp(-y_pred));
return y_pred;
}
// Stochastic gradient descent
void sgd( int m, int n, double *X, double *y, double *coef, double *intercept, double eta, int max_iter, int fit_intercept, int random_seed )
{
double *gradient_coef, *X_i;
double y_i, y_pred, resid;
int idx;
double gradient_intercept = 0.0, alpha = 1.0, beta = 1.0;
X_i = (double *) malloc (n * sizeof(double));
gradient_coef = (double *) malloc (n * sizeof(double));
for ( int i = 0; i < n; i++ )
{
coef[i] = 0.0;
gradient_coef[i] = 0.0;
}
*intercept = 0.0;
srand(random_seed);
for ( int epoch = 0; epoch < max_iter; epoch++ )
{
for ( int run = 0; run < m; run++ )
{
// Randomly sample an observation
idx = rand() % m;
for ( int i = 0; i < n; i++ )
{
X_i[i] = X[n*idx+i];
}
y_i = y[idx];
// Compute y_hat
y_pred = sigmoid( n, alpha, X_i, coef, beta, *intercept );
resid = -(y_i - y_pred);
// Compute gradients and adjust weights
for (int i = 0; i < n; i++)
{
gradient_coef[i] = X_i[i] * resid;
coef[i] -= eta * gradient_coef[i];
}
if ( fit_intercept == 1 )
{
*intercept -= eta * resid;
}
}
}
}
int main(void)
{
double *X, *y, *coef, *y_pred;
double intercept;
double eta = 0.05;
double alpha = 1.0, beta = 1.0;
long m = 50000;
long n = 150;
int max_iter = 20;
long class_0 = (long)(3.0 / 4.0 * (double)m);
double pct_class_1 = 0.0;
clock_t test_start;
clock_t test_end;
double test_time;
printf("Constructing variables...\n");
X = (double *) malloc (m * n * sizeof(double));
y = (double *) malloc (m * sizeof(double));
y_pred = (double *) malloc (m * sizeof(double));
coef = (double *) malloc (n * sizeof(double));
// Initialize classes
for (int i = 0; i < m; i++)
{
if (i < class_0)
{
y[i] = 0.0;
}
else
{
y[i] = 1.0;
}
}
// Initialize observation features
for (int i = 0; i < m; i++)
{
if (i < class_0)
{
X[n*i] = 1.0 / (double)m;
}
else
{
X[n*i] = (double)i / (double)m;
}
X[n*i + 1] = (double)i / (double)m;
for (int j = 2; j < n; j++)
{
X[n*i + j] = (double)(rand() % 100) / 100.0;
}
}
// Fit weights
printf("Running SGD...\n");
test_start = clock();
sgd( m, n, X, y, coef, &intercept, eta, max_iter, 1, 42 );
test_end = clock();
test_time = (double)(test_end - test_start) / CLOCKS_PER_SEC;
printf("Time taken: %f\n", test_time);
// Compute y_hat and share of observations predicted as class 1
printf("Making predictions...\n");
for ( int i = 0; i < m; i++ )
{
y_pred[i] = sigmoid( n, alpha, &X[i*n], coef, beta, intercept );
}
printf("Printing results...\n");
for ( int i = 0; i < m; i++ )
{
//printf("%f\n", y_pred[i]);
if (y_pred[i] > 0.5)
{
pct_class_1 += 1.0;
}
// Troubleshooting print
if (i < 10 || i > m - 10)
{
printf("%g\n", y_pred[i]);
}
}
printf("Percent class 1: %f", pct_class_1 / (double)m);
return 0;
}
For reference, here is my (presumably) equivalent Python code, which returns the correct percent of identified classes at more than 50,000 observations:
import numpy as np
import time
def sigmoid(x):
return 1 / (1 + np.exp(-x))
class LogisticRegressor:
def __init__(self, eta, init_runs, fit_intercept=True):
self.eta = eta
self.init_runs = init_runs
self.fit_intercept = fit_intercept
def fit(self, x, y):
m, n = x.shape
self.coef = np.zeros((n, 1))
self.intercept = np.zeros((1, 1))
for epoch in range(self.init_runs):
for run in range(m):
idx = np.random.randint(0, m)
x_i = x[idx:idx+1, :]
y_i = y[idx]
y_pred_i = sigmoid(x_i.dot(self.coef) + self.intercept)
gradient_w = -(x_i.T * (y_i - y_pred_i))
self.coef -= self.eta * gradient_w
if self.fit_intercept:
gradient_b = -(y_i - y_pred_i)
self.intercept -= self.eta * gradient_b
def predict_proba(self, x):
m, n = x.shape
y_pred = np.ones((m, 2))
y_pred[:,1:2] = sigmoid(x.dot(self.coef) + self.intercept)
y_pred[:,0:1] -= y_pred[:,1:2]
return y_pred
def predict(self, x):
return np.round(sigmoid(x.dot(self.coef) + self.intercept))
m = 50000
n = 150
class1 = int(3.0 / 4.0 * m)
X = np.random.rand(m, n)
y = np.zeros((m, 1))
for obs in range(m):
if obs < class1:
continue
else:
y[obs,0] = 1
for obs in range(m):
if obs < class1:
X[obs, 0] = 1.0 / float(m)
else:
X[obs, 0] = float(obs) / float(m)
X[obs, 1] = float(obs) / float(m)
logit = LogisticRegressor(0.05, 20)
start_time = time.time()
logit.fit(X, y)
end_time = time.time()
print(round(end_time - start_time, 2))
y_pred = logit.predict(X)
print("Percent:", y_pred.sum() / len(y_pred))
The issue is here:
// Randomly sample an observation
idx = rand() % m;
... in light of the fact that the OP's RAND_MAX is 32767. This is exacerbated by the fact that all of the class 0 observations are at the end.
All samples will be drawn from the first 32768 observations, and when the total number of observations is greater than that, the proportion of class 0 observations among those that can be sampled is less than 0.25. At 43691 total observations, there are no class 0 observations among those that can be sampled.
As a secondary issue, rand() % m does not yield a wholly uniform distribution if m does not evenly divide RAND_MAX + 1, though the effect of this issue will be much more subtle.
Bottom line: you need a better random number generator.
At minimum, you could consider combining the bits from two calls to rand() to yield an integer with sufficient range, but you might want to consider getting a third-party generator. There are several available.
Note: OP reports "m=50,000 observations with n=150 features.", so perhaps this is not the issue for OP, but I'll leave this answer up for reference when OP tries larger tasks.
A potential issue:
long overflow
m * n * sizeof(double) risks overflow when long is 32-bit and m*n > LONG_MAX (or about 46,341 if m, n are the same).
OP does report
A first step is to perform the multiplication using size_t math where we gain at least 1 more bit in the calculation.
// m * n * sizeof(double)
sizeof(double) * m * n
Yet unless OP's size_t is more than 32-bit, we still have trouble.
IAC, I recommend to use size_t for array sizing and indexing.
Check allocations for failure too.
Since RAND_MAX may be too small and array indexing should be done using size_t math, consider a helper function to generate a random index over the entire size_t range.
// idx = rand() % m;
size_t idx = rand_size_t() % (size_t)m;
If stuck with the standard rand(), below is a helper function to extend its range as needed.
It uses the real nifty IMAX_BITS(m).
#include <assert.h>
#include <limits.h>
#include <stdint.h>
#include <stdlib.h>
// https://stackoverflow.com/a/4589384/2410359
/* Number of bits in inttype_MAX, or in any (1<<k)-1 where 0 <= k < 2040 */
#define IMAX_BITS(m) ((m)/((m)%255+1) / 255%255*8 + 7-86/((m)%255+12))
// Test that RAND_MAX is a power of 2 minus 1
_Static_assert((RAND_MAX & 1) && ((RAND_MAX/2 + 1) & (RAND_MAX/2)) == 0, "RAND_MAX is not a Mersenne number");
#define RAND_MAX_WIDTH (IMAX_BITS(RAND_MAX))
#define SIZE_MAX_WIDTH (IMAX_BITS(SIZE_MAX))
size_t rand_size_t(void) {
size_t index = (size_t) rand();
for (unsigned i = RAND_MAX_WIDTH; i < SIZE_MAX_WIDTH; i += RAND_MAX_WIDTH) {
index <<= RAND_MAX_WIDTH;
index ^= (size_t) rand();
}
return index;
}
Further considerations can replace the rand_size_t() % (size_t)m with a more uniform distribution.
As has been determined elsewhere, the problem is due to the implementation's RAND_MAX value being too small.
Assuming 32-bit ints, a slightly better PRNG function can be implemented in the code, such as this C implementation of the minstd_rand() function from C++:
#define MINSTD_RAND_MAX 2147483646
// Code assumes `int` is at least 32 bits wide.
static unsigned int minstd_seed = 1;
static void minstd_srand(unsigned int seed)
{
seed %= 2147483647;
// zero seed is bad!
minstd_seed = seed ? seed : 1;
}
static int minstd_rand(void)
{
minstd_seed = (unsigned long long)minstd_seed * 48271 % 2147483647;
return (int)minstd_seed;
}
Another problem is that expressions of the form rand() % m produce a biased result when m does not divide (unsigned int)RAND_MAX + 1. Here is an unbiased function that returns a random integer from 0 to le inclusive, making use of the minstd_rand() function defined earlier:
static int minstd_rand_max(int le)
{
int r;
if (le < 0)
{
r = le;
}
else if (le >= MINSTD_RAND_MAX)
{
r = minstd_rand();
}
else
{
int rm = MINSTD_RAND_MAX - le + MINSTD_RAND_MAX % (le + 1);
while ((r = minstd_rand()) > rm)
{
}
r /= (rm / (le + 1) + 1);
}
return r;
}
(Actually, it does still have a very small bias because minstd_rand() will never return 0.)
For example, replace rand() % 100 with minstd_rand_max(99), and replace rand() % m with minstd_rand_max(m - 1). Also replace srand(random_seed) with minstd_srand(random_seed).
The function should take the address of the integer and modify it by inserting zeros between its digits. For example:
insert_zeros(3) //3
insert_zeros(39) //309
insert_zeros(397) //30907
insert_zeros(3976) //3090706
insert_zeros(39765) //309070605
My code:
#include <stdio.h>
#include <math.h>
void insert_zeros(int* num);
int main() {
int num;
printf("Enter a number:");
scanf("%d", num);
insert_zeros(&num);
printf("Number after inserting zeros: %d", num);
return 0;
}
void insert_zeros(int* num){
int count = 0;
int tmp = *num;
//Count the number of digits in the number
while(tmp != 0){
tmp /= 10;
count++;
}
//calculating the coefficient by which I will divide the number to get its digits one by one
int divider = (int)pow(10, count-1);
int multiplier;
tmp = *num;
*num = 0;
/*
The point at which I'm stuck
Here I tried to calculate the degree for the number 10
(my thought process and calculations are provided below)
*/
(count >= 3)? count += (count/2): count;
//the main loop of assembling the required number
while (count >= 0){
multiplier = (int)pow(10, count); //calculating a multiplier
*num += (tmp / divider) * multiplier; //assembling the required number
tmp %= divider; //removing the first digit of the number
divider /= 10; //decreasing divider
count -= 2; //decreasing the counter,
//which is also a power of the multiplier (witch is 10)
}
}
My idea consists of the following formula:
For number "3" I shold get "30" and it will be:
30 = (3 * 10^1) - the power is a counter for number "3" that equals 1.
For number "39" it will be "309":
309 = (3 * 10^2) + (9 * 10^1)
For number "397" it will be "30907":
30907 = (3 * 10^4) + (9 * 10^2) + (7 * 10^0)
For number "3976" it will be "3090706":
3090706 = (3 * 10^6) + (9 * 10^4) + (7 * 10^2) + (6 * 10^0) - with each iteration power is decreasing by 2
For number "39765" it will be "309070605":
309070605 = (3 * 10^8) + (9 * 10^6) + (7 * 10^4) + (6 * 10^2) + (5 * 10^0)
And so on...
For a 3-digit number, the start power should be 4, for a 4-digit number power should be 6, for a 5-digit it should be 8, for 6-digit it should be 10, etc.
That algorithm works until it takes a 5-digit number. It outputs a number like "30907060" with an extra "0" at the end.
And the main problem is in that piece of code (count >= 3)? count += (count/2): count;, where I tried to calculate the right power for the first iterating through the loop. It should give the right number to which will be added all the following numbers. But it only works until it gets a 5-digit number.
To be honest, so far I don't really understand how it can be realized. I would be very grateful if someone could explain how this can be done.
As noted in comments, your use of scanf is incorrect. You need to pass a pointer as the second argument.
#include <stdio.h>
#include <math.h>
int main(void) {
int num;
scanf("%d", &num);
int num2 = 0;
int power = 0;
while (num > 0) {
num2 += (num % 10) * (int)pow(10, power);
num /= 10;
power += 2;
}
printf("%d\n", num2);
return 0;
}
There's an easy recursive formula for inserting zeros: IZ(n) = 100*IZ(n/10) + n%10.
That gives a very concise solution -- here the test cases are more code than the actual function itself.
#include <stdio.h>
#include <stdint.h>
uint64_t insert_zeros(uint64_t n) {
return n ? (100 * insert_zeros(n / 10) + n % 10) : 0;
}
int main(int argc, char **argv) {
int tc[] = {1, 12, 123, 9854, 12345, 123450};
for (int i = 0; i < sizeof(tc)/sizeof(*tc); i++) {
printf("%d -> %lu\n", tc[i], insert_zeros(tc[i]));
}
}
Output:
1 -> 1
12 -> 102
123 -> 10203
9854 -> 9080504
12345 -> 102030405
123450 -> 10203040500
Adapting some code just posted for another of these silly exercises:
int main() {
int v1 = 12345; // I don't like rekeying data. Here's the 'seed' value.
printf( "Using %d as input\n", v1 );
int stack[8] = { 0 }, spCnt = -1;
// Peel off each 'digit' right-to-left, pushing onto a stack
while( v1 )
stack[ ++spCnt ] = v1%10, v1 /= 10;
if( spCnt == 0 ) // Special case for single digit seed.
v1 = stack[ spCnt ] * 10;
else
// multiply value sofar by 100, and add next digit popped from stack.
while( spCnt >= 0 )
v1 = v1 * 100 + stack[ spCnt-- ];
printf( "%d\n", v1 );
return 0;
}
There's a ceiling to how big a decimal value can be stored in an int. If you want to start to play with strings of digits, that is another matter entirely.
EDIT: If this were in Java, this would be a solution, but the problem is in C, which I'm not sure if this can convert to C.
This may be a lot easier if you first convert the integer to a string, then use a for loop to add the zeros, then afterward reconvert to an integer. Example:
int insert_zeros(int num) {
String numString = Integer.toString(num);
String newString = "";
int numStringLength = numString.length();
for (int i = 0; i < numStringLength; i++) {
newString += numString[i];
// Only add a 0 if it's not the last digit (with exception of 1st digit)
if (i < numStringLength - 1 || i == 0) newString += '0';
}
return Integer.parseInt(newString);
}
I think this should give you your desired effect. It's been a little bit since I've worked with Java (I'm currently doing JavaScript), so I hope there's no syntax errors, but the logic should all be correct.
I took this example from a video in youtube, where a sen(x) is resolved using this approach, I just changed the formula to an arcsin(x).
I am trying to figure out why this code is not working as expected:
#include <stdio.h>
//#include <float.h>
#include <math.h>
int factorial(int x)
{
int fac = 1;
while (x!=0)
{
fac= fac*x;
x--;
}
return fac;
}
int main()
{
int i;
float x, sum, divisor = 1, dividendo = 1, temp;
printf("enter the value of X(in degrees): ");
scanf("%f", &x);
//x = x*3.141592/180;
sum = 0;
for(i=1; ; i++)
{
divisor = (factorial(2*i));
dividendo = (pow(4,(i)) * pow(factorial(i),2)*(2*i+1));
temp = (divisor/dividendo)*pow(x,(2*i+1));
if (temp < FLT_EPSILON)
break;
sum = sum + temp;
printf("our result = %f\n", sum);
}
printf("reference = %f\n", asin(x));
// printf("our result = %f\n", sum);
}
Below is the result, and as you may see the output I get is far below from what is expected. the result printed comes from every iteration, and the reference is just to make a comparison between both outcomes.
enter the value of X(in degrees): 1
our result = 0.166667
our result = 0.241667
our result = 0.286310
our result = 0.316691
our result = 0.339064
our result = 0.356416
our result = 0.356621
our result = 0.356622
our result = 0.356622
our result = 0.356622
**reference = 1.570796**
The Taylor series for arscin is
arcsin(x) = sum n = 0; inf; (2*n)! / (4**n * (n!)**2 * (2*n + 1)) * x**(2*n + 1)
where ** is the power operator and ! denotes a factorial. As was already noted in comments, an int can represent factorials up to 12! only; a 64-bit long can represent factorials up to 20!. Since you use (2*n)!, these limits will be reached quickly.
A better solution to this problem is not to calculate every term separately as given in the formula, but to evolve the term by calculating it from the previous term:
term(n + 1) = fact * term(n)
Since each term is one large factor, you can do that for each subfactor:
2*(n + 1))! = (2*n)! * (2*n + 1) * (2*n + 2)
4**(n + 1) = 4**n * 4
((n + 1)!)**2 = n!**2 * (n + 1)**2
2*(n + 1) + 1 = 2*(n + 1) * (2*n + 3) / (2*n + 1)
x**(2*(n + 1) + 1) = x**(2*n + 1) * x**2
Putting all this together:
float res = x; // first term ...
float fact = x; // ...equals the first factor
for (int n = 0; n < nMax; n++) {
float old = res;
// calculate term(n + 1) as per the formulas above
fact *= (2*n + 1) * (2*n + 2);
fact /= 4.0 * (n + 1)*(n + 1) * (2*n + 3);
fact *= x * x * (2*n + 1);
res += fact;
if (res == old) break;
printf("[%d] %f\n", n, res);
}
printf("ref %f\n", asin(x));
(FLT_EPSILON is the granularity of floating-point numbers near 1.0. The terms are converging to zero, where the granularity is finer. I've tested whether adding the new term doesn't change the sum as a convergence criterion with a fixed maximum number of iterations, nMax.)
This series does not converge well near ±1, where the slope of the function approaches infinity. By Stirling's approximation the nth term approximately equals 1/((2*n+1)*√(πn)), which converges very slowly. For the nth term to be less than FLT_EPSILON, n needs to be greater than 26000. Unfortunately, with that many summations, the finite precision of floating point numbers would prevent the series to converge to the correct answer.
Let's say I've been given two integers a, b where a is a positive integer and is smaller than b. I have to find an efficient algorithm that's going to give me the sum of number of base2 digits (number of bits) over the interval [a, b]. For example, in the interval [0, 4] the sum of digits is equal to 9 because 0 = 1 digit, 1 = 1 digit, 2 = 2 digits, 3 = 2 digits and 4 = 3 digits.
My program is capable of calculating this number by using a loop but I'm looking for something more efficient for large numbers. Here are the snippets of my code just to give you an idea:
int numberOfBits(int i) {
if(i == 0) {
return 1;
}
else {
return (int) log2(i) + 1;
}
}
The function above is for calculating the number of digits of one number in the interval.
The code below shows you how I use it in my main function.
for(i = a; i <= b; i++) {
l = l + numberOfBits(i);
}
printf("Digits: %d\n", l);
Ideally I should be able to get the number of digits by using the two values of my interval and using some special algorithm to do that.
Try this code, i think it gives you what you are needing to calculate the binaries:
int bit(int x)
{
if(!x) return 1;
else
{
int i;
for(i = 0; x; i++, x >>= 1);
return i;
}
}
The main thing to understand here is that the number of digits used to represent a number in binary increases by one with each power of two:
+--------------+---------------+
| number range | binary digits |
+==============+===============+
| 0 - 1 | 1 |
+--------------+---------------+
| 2 - 3 | 2 |
+--------------+---------------+
| 4 - 7 | 3 |
+--------------+---------------+
| 8 - 15 | 4 |
+--------------+---------------+
| 16 - 31 | 5 |
+--------------+---------------+
| 32 - 63 | 6 |
+--------------+---------------+
| ... | ... |
A trivial improvement over your brute force algorithm would then be to figure out how many times this number of digits has increased between the two numbers passed in (given by the base two logarithm) and add up the digits by multiplying the count of numbers that can be represented by the given number of digits (given by the power of two) with the number of digits.
A naive implementation of this algorithm is:
int digits_sum_seq(int a, int b)
{
int sum = 0;
int i = 0;
int log2b = b <= 0 ? 1 : floor(log2(b));
int log2a = a <= 0 ? 1 : floor(log2(a)) + 1;
sum += (pow(2, log2a) - a) * (log2a);
for (i = log2b; i > log2a; i--)
sum += pow(2, i - 1) * i;
sum += (b - pow(2, log2b) + 1) * (log2b + 1);
return sum;
}
It can then be improved by the more efficient versions of the log and pow functions seen in the other answers.
First, we can improve the speed of log2, but that only gives us a fixed factor speed-up and doesn't change the scaling.
Faster log2 adapted from: https://graphics.stanford.edu/~seander/bithacks.html#IntegerLogLookup
The lookup table method takes only about 7 operations to find the log
of a 32-bit value. If extended for 64-bit quantities, it would take
roughly 9 operations. Another operation can be trimmed off by using
four tables, with the possible additions incorporated into each. Using
int table elements may be faster, depending on your architecture.
Second, we must re-think the algorithm. If you know that numbers between N and M have the same number of digits, would you add them up one by one or would you rather do (M-N+1)*numDigits?
But if we have a range where multiple numbers appear what do we do? Let's just find the intervals of same digits, and add sums of those intervals. Implemented below. I think that my findEndLimit could be further optimized with a lookup table.
Code
#include <stdio.h>
#include <limits.h>
#include <time.h>
unsigned int fastLog2(unsigned int v)
{
static const char LogTable256[256] =
{
#define LT(n) n, n, n, n, n, n, n, n, n, n, n, n, n, n, n, n
-1, 0, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3, 3,
LT(4), LT(5), LT(5), LT(6), LT(6), LT(6), LT(6),
LT(7), LT(7), LT(7), LT(7), LT(7), LT(7), LT(7), LT(7)
};
register unsigned int t, tt; // temporaries
if (tt = v >> 16)
{
return (t = tt >> 8) ? 24 + LogTable256[t] : 16 + LogTable256[tt];
}
else
{
return (t = v >> 8) ? 8 + LogTable256[t] : LogTable256[v];
}
}
unsigned int numberOfBits(unsigned int i)
{
if (i == 0) {
return 1;
}
else {
return fastLog2(i) + 1;
}
}
unsigned int findEndLimit(unsigned int sx, unsigned int ex)
{
unsigned int sy = numberOfBits(sx);
unsigned int ey = numberOfBits(ex);
unsigned int mx;
unsigned int my;
if (sy == ey) // this also means sx == ex
return ex;
// assumes sy < ey
mx = (ex - sx) / 2 + sx; // will eq. sx for sx + 1 == ex
my = numberOfBits(mx);
while (ex - sx != 1) {
mx = (ex - sx) / 2 + sx; // will eq. sx for sx + 1 == ex
my = numberOfBits(mx);
if (my == ey) {
ex = mx;
ey = numberOfBits(ex);
}
else {
sx = mx;
sy = numberOfBits(sx);
}
}
return sx+1;
}
int main(void)
{
unsigned int a, b, m;
unsigned long l;
clock_t start, end;
l = 0;
a = 0;
b = UINT_MAX;
start = clock();
unsigned int i;
for (i = a; i < b; ++i) {
l += numberOfBits(i);
}
if (i == b) {
l += numberOfBits(i);
}
end = clock();
printf("Naive\n");
printf("Digits: %ld; Time: %fs\n",l, ((double)(end-start))/CLOCKS_PER_SEC);
l=0;
start = clock();
do {
m = findEndLimit(a, b);
l += (b-m + 1) * (unsigned long)numberOfBits(b);
b = m-1;
} while (b > a);
l += (b-a+1) * (unsigned long)numberOfBits(b);
end = clock();
printf("Binary search\n");
printf("Digits: %ld; Time: %fs\n",l, ((double)(end-start))/CLOCKS_PER_SEC);
}
Output
From 0 to UINT_MAX
$ ./main
Naive
Digits: 133143986178; Time: 25.722492s
Binary search
Digits: 133143986178; Time: 0.000025s
My findEndLimit can take long time in some edge cases:
From UINT_MAX/16+1 to UINT_MAX/8
$ ./main
Naive
Digits: 7784628224; Time: 1.651067s
Binary search
Digits: 7784628224; Time: 4.921520s
Conceptually, you would need to split the task to two subproblems -
1) find the sum of digits from 0..M, and from 0..N, then subtract.
2) find the floor(log2(x)), because eg for the number 77 the numbers 64,65,...77 all have 6 digits, the next 32 have 5 digits, the next 16 have 4 digits and so on, which makes a geometric progression.
Thus:
int digits(int a) {
if (a == 0) return 1; // should digits(0) be 0 or 1 ?
int b=(int)floor(log2(a)); // use any all-integer calculation hack
int sum = 1 + (b+1) * (a- (1<<b) +1); // added 1, due to digits(0)==1
while (--b)
sum += (b + 1) << b; // shortcut for (b + 1) * (1 << b);
return sum;
}
int digits_range(int a, int b) {
if (a <= 0 || b <= 0) return -1; // formulas work for strictly positive numbers
return digits(b)-digits(a-1);
}
As efficiency depends on the tools available, one approach would be doing it "analog":
#include <stdlib.h>
#include <stdio.h>
#include <math.h>
unsigned long long pow2sum_min(unsigned long long n, long long unsigned m)
{
if (m >= n)
{
return 1;
}
--n;
return (2ULL << n) + pow2sum_min(n, m);
}
#define LN(x) (log2(x)/log2(M_E))
int main(int argc, char** argv)
{
if (2 >= argc)
{
fprintf(stderr, "%s a b\n", argv[0]);
exit(EXIT_FAILURE);
}
long a = atol(argv[1]), b = atol(argv[2]);
if (0L >= a || 0L >= b || b < a)
{
puts("Na ...!");
exit(EXIT_FAILURE);
}
/* Expand intevall to cover full dimensions: */
unsigned long long a_c = pow(2, floor(log2(a)));
unsigned long long b_c = pow(2, floor(log2(b+1)) + 1);
double log2_a_c = log2(a_c);
double log2_b_c = log2(b_c);
unsigned long p2s = pow2sum_min(log2_b_c, log2_a_c) - 1;
/* Integral log2(x) between a_c and b_c: */
double A = ((b_c * (LN(b_c) - 1))
- (a_c * (LN(a_c) - 1)))/LN(2)
+ (b+1 - a);
/* "Integer"-integral - integral of log2(x)'s inverse function (2**x) between log(a_c) and log(b_c): */
double D = p2s - (b_c - a_c)/LN(2);
/* Corrective from a_c/b_c to a/b : */
double C = (log2_b_c - 1)*(b_c - (b+1)) + log2_a_c*(a - a_c);
printf("Total used digits: %lld\n", (long long) ((A - D - C) +.5));
}
:-)
The main thing here is the number and kind of iterations done.
Number is
log(floor(b_c)) - log(floor(a_c))
times
doing one
n - 1 /* Integer decrement */
2**n + s /* One bit-shift and one integer addition */
for each iteration.
Here's an entirely look-up based approach. You don't even need the log2 :)
Algorithm
First we precompute interval limits where the number of bits would change and create a lookup table. In other words we create an array limits[2^n], where limits[i] gives us the biggest integer that can be represented with (i+1) bits. Our array is then {1, 3, 7, ..., 2^n-1}.
Then, when we want to determine the sum of bits for our range, we must first match our range limits a and b with the smallest index for which a <= limits[i] and b <= limits[j] holds, which will then tell us that we need (i+1) bits to represent a, and (j+1) bits to represent b.
If the indexes are the same, then the result is simply (b-a+1)*(i+1), otherwise we must separately get the number of bits from our value to the edge of same number of bits interval, and add up total number of bits for each interval between as well. In any case, simple arithmetic.
Code
#include <stdio.h>
#include <limits.h>
#include <time.h>
unsigned long bitsnumsum(unsigned int a, unsigned int b)
{
// generate lookup table
// limits[i] is the max. number we can represent with (i+1) bits
static const unsigned int limits[32] =
{
#define LTN(n) n*2u-1, n*4u-1, n*8u-1, n*16u-1, n*32u-1, n*64u-1, n*128u-1, n*256u-1
LTN(1),
LTN(256),
LTN(256*256),
LTN(256*256*256)
};
// make it work for any order of arguments
if (b < a) {
unsigned int c = a;
a = b;
b = c;
}
// find interval of a
unsigned int i = 0;
while (a > limits[i]) {
++i;
}
// find interval of b
unsigned int j = i;
while (b > limits[j]) {
++j;
}
// add it all up
unsigned long sum = 0;
if (i == j) {
// a and b in the same range
// conveniently, this also deals with j == 0
// so no danger to do [j-1] below
return (i+1) * (unsigned long)(b - a + 1);
}
else {
// add sum of digits in range [a, limits[i]]
sum += (i+1) * (unsigned long)(limits[i] - a + 1);
// add sum of digits in range [limits[j], b]
sum += (j+1) * (unsigned long)(b - limits[j-1]);
// add sum of digits in range [limits[i], limits[j]]
for (++i; i<j; ++i) {
sum += (i+1) * (unsigned long)(limits[i] - limits[i-1]);
}
return sum;
}
}
int main(void)
{
clock_t start, end;
unsigned int a=0, b=UINT_MAX;
start = clock();
printf("Sum of binary digits for numbers in range "
"[%u, %u]: %lu\n", a, b, bitsnumsum(a, b));
end = clock();
printf("Time: %fs\n", ((double)(end-start))/CLOCKS_PER_SEC);
}
Output
$ ./lookup
Sum of binary digits for numbers in range [0, 4294967295]: 133143986178
Time: 0.000282s
Algorithm
The main idea is to find the n2 = log2(x) rounded down. That is the number of digits in x. Let pow2 = 1 << n2. n2 * (pow2 - x + 1) is the number of digits in the values [x...pow2]. Now find the sun of digits in the powers of 2 from 1 to n2-1
Code
I am certain various simplifications can be made.
Untested code. Will review later.
// Let us use unsigned for everything.
unsigned ulog2(unsigned value) {
unsigned result = 0;
if (0xFFFF0000u & value) {
value >>= 16; result += 16;
}
if (0xFF00u & value) {
value >>= 8; result += 8;
}
if (0xF0u & value) {
value >>= 4; result += 4;
}
if (0xCu & value) {
value >>= 2; result += 2;
}
if (0x2 & value) {
value >>= 1; result += 1;
}
return result;
}
unsigned bit_count_helper(unsigned x) {
if (x == 0) {
return 1;
}
unsigned n2 = ulog2(x);
unsigned pow2 = 1u << n;
unsigned sum = n2 * (pow2 - x + 1u); // value from pow2 to x
while (n2 > 0) {
// ... + 5*16 + 4*8 + 3*4 + 2*2 + 1*1
pow2 /= 2;
sum += n2 * pow2;
}
return sum;
}
unsigned bit_count(unsigned a, unsigned b) {
assert(a < b);
return bit_count_helper(b - 1) - bit_count_helper(a);
}
For this problem your solution is the simplest, the one called "naive" where you look for every element in the sequence or in your case interval for check something or execute operations.
Naive Algorithm
Assuming that a and b are positive integers with b greater than a let's call the dimension/size of the interval [a,b], n = (b-a).
Having our number of elements n and using some notations of algorithms (like big-O notation link), the worst case cost is O(n*(numberOfBits_cost)).
From this we can see that we can speed up our algorithm by using a faster algorithm for computing numberOfBits() or we need to find a way to not look at every element of the interval that costs us n operations.
Intuition
Now looking at a possible interval [6,14] you can see that for 6 and 7 we need 3 digits, with 4 need for 8,9,10,11,12,13,14. This results in calling numberOfBits() for every number that use the same number of digits to be represented, while the following multiplication operation would be faster:
(number_in_subinterval)*digitsForThisInterval
((14-8)+1)*4 = 28
((7-6)+1)*3 = 6
So we reduced the looping on 9 elements with 9 operations to only 2.
So writing a function that use this intuition will give us a more efficient in time, not necessarily in memory, algorithm. Using your numberOfBits() function I have created this solution:
int intuitionSol(int a, int b){
int digitsForA = numberOfBits(a);
int digitsForB = numberOfBits(b);
if(digitsForA != digitsForB){
//because a or b can be that isn't the first or last element of the
// interval that a specific number of digit can rappresent there is a need
// to execute some correction operation before on a and b
int tmp = pow(2,digitsForA) - a;
int result = tmp*digitsForA; //will containt the final result that will be returned
int i;
for(i = digitsForA + 1; i < digitsForB; i++){
int interval_elements = pow(2,i) - pow(2,i-1);
result = result + ((interval_elements) * i);
//printf("NumOfElem: %i for %i digits; sum:= %i\n", interval_elements, i, result);
}
int tmp1 = ((b + 1) - pow(2,digitsForB-1));
result = result + tmp1*digitsForB;
return result;
}
else {
int elements = (b - a) + 1;
return elements * digitsForA; // or digitsForB
}
}
Let's look at the cost, this algorithm costs is the cost of doing correction operation on a and b plus the most expensive one that of the for-loop. In my solution however I'm not looping over all elements but only on numberOfBits(b)-numberOfBits(a) that in the worst case, when [0,n], become log(n)-1 thats equivalent to O(log n).
To resume we passed from a linear operations cost O(n) to a logartmic one O(log n) in the worst case. Look on this diagram the diferinces between the two.
Note
When I talk about interval or sub-interval I refer to the interval of elements that use the same number of digits to represent the number in binary.
Following there are some output of my tests with the last one that shows the difference:
Considered interval is [0,4]
YourSol: 9 in time: 0.000015s
IntuitionSol: 9 in time: 0.000007s
Considered interval is [0,0]
YourSol: 1 in time: 0.000005s
IntuitionSol: 1 in time: 0.000005s
Considered interval is [4,7]
YourSol: 12 in time: 0.000016s
IntuitionSol: 12 in time: 0.000005s
Considered interval is [2,123456]
YourSol: 1967697 in time: 0.005010s
IntuitionSol: 1967697 in time: 0.000015s
I am making simple calculator and it is e^x function part.
it works for positive number, but it doesn't for negative x.
How can I make it works for negative x too?`
double calculateEx(double x) {
double beforeResult = 1, afterResult = 1, term = 1, error = 1, i = 1, j;
while (error > 0.001) {
afterResult = beforeResult;
for (j = 1; j <= i; j++) {
term *= x;
}
term /= fact(i);
afterResult += term;
error = (afterResult - beforeResult) / afterResult;
if (error < 0) error * -1;
error *= 100;
beforeResult = afterResult;
term = 1;
i++;
}
return beforeResult;
}
double fact (double num) {
int i, j;
double total = 1;
for (i = 2; i <= num; i++) {
total = total * i;
}
return total;
}
When computing exponent via Taylor serie
exp(x) = 1 + x / 1 + x**2/2! + ... + x**n/n!
you don't want any factorials, please, notice that if n-1th term is
t(n-1) = x**(n-1)/(n-1)!
then
t(n) = x**n/n! = t(n-1) * x / n;
That's why all you have to implement is:
double calculateEx(double x) {
double term = 1.0;
double result = term;
/*
the only trick is that term can be positive as well as negative;
we should either use abs in any implementation or putr two conditions
*/
for (int n = 1; term > 0.001 || term < -0.001; ++n) {
term = term * x / n;
result += term;
}
return result;
}
OK, as I wrote in a comment above, I'd use <math.h> if at all possible, but since you asked the question:
To make it work with negative numbers, if x is negative, consider what happens if you negate it.
You can get rid of the factorial function by storing a table of factorials. You won't need that many elements.