move some functions from scratchpad => squiggle.c, reorg

This commit is contained in:
NunoSempere 2023-07-16 21:26:33 +02:00
parent 68e7730f24
commit ee9ed34287
5 changed files with 232 additions and 233 deletions

View File

@ -9,7 +9,7 @@ CC=gcc # required for nested functions
# CC=tcc # <= faster compilation # CC=tcc # <= faster compilation
# Main file # Main file
SRC=scratchpad.c SRC=scratchpad.c ../squiggle.c
OUTPUT=./scratchpad OUTPUT=./scratchpad
## Dependencies ## Dependencies

Binary file not shown.

View File

@ -1,33 +1,13 @@
#include <float.h> // FLT_MAX, FLT_MIN
#include <limits.h> // INT_MAX
#include <math.h> // erf, sqrt #include <math.h> // erf, sqrt
#include <stdint.h> #include <stdint.h>
#include <stdio.h> #include <stdio.h>
#include <stdlib.h> #include <stdlib.h>
// #include <sys/types.h>
#include <time.h> #include <time.h>
#include "../squiggle.h"
#define EXIT_ON_ERROR 0
#define MAX_ERROR_LENGTH 500
#define PROCESS_ERROR(...) \
do { \
if (EXIT_ON_ERROR) { \
printf("@, in %s (%d)", __FILE__, __LINE__); \
exit(1); \
} else { \
char error_msg[MAX_ERROR_LENGTH]; \
snprintf(error_msg, MAX_ERROR_LENGTH, "@, in %s (%d)", __FILE__, __LINE__); \
struct box error = { .empty = 1, .error_msg = error_msg }; \
return error; \
} \
} while (0)
#define NUM_SAMPLES 1000000 #define NUM_SAMPLES 1000000
#define STOP_BETA 1.0e-8
struct box { #define TINY_BETA 1.0e-30
int empty;
float content;
char* error_msg;
};
// Example cdf // Example cdf
float cdf_uniform_0_1(float x) float cdf_uniform_0_1(float x)
@ -59,16 +39,10 @@ float cdf_normal_0_1(float x)
return 0.5 * (1 + erf((x - mean) / (std * sqrt(2)))); // erf from math.h return 0.5 * (1 + erf((x - mean) / (std * sqrt(2)))); // erf from math.h
} }
// [x] to do: add beta.
// [x] for the cdf, use this incomplete beta function implementation, based on continuous fractions:
// <https://codeplea.com/incomplete-beta-function-c>
// <https://github.com/codeplea/incbeta>
#define STOP_BETA 1.0e-8
#define TINY_BETA 1.0e-30
struct box incbeta(float a, float b, float x) struct box incbeta(float a, float b, float x)
{ {
// Descended from <https://github.com/codeplea/incbeta/blob/master/incbeta.c>, // Descended from <https://github.com/codeplea/incbeta/blob/master/incbeta.c>,
// <https://codeplea.com/incomplete-beta-function-c>
// but modified to return a box struct and floats instead of doubles. // but modified to return a box struct and floats instead of doubles.
// [ ] to do: add attribution in README // [ ] to do: add attribution in README
// Original code under this license: // Original code under this license:
@ -174,200 +148,6 @@ struct box cdf_beta(float x)
} }
} }
// Inverse cdf at point
// Two versions of this function:
// - raw, dealing with cdfs that return floats
// - box, dealing with cdfs that return a box.
// Inverse cdf
struct box inverse_cdf_float(float cdf(float), float p)
{
// given a cdf: [-Inf, Inf] => [0,1]
// returns a box with either
// x such that cdf(x) = p
// or an error
// if EXIT_ON_ERROR is set to 1, it exits instead of providing an error
float low = -1.0;
float high = 1.0;
// 1. Make sure that cdf(low) < p < cdf(high)
int interval_found = 0;
while ((!interval_found) && (low > -FLT_MAX / 4) && (high < FLT_MAX / 4)) {
// ^ Using FLT_MIN and FLT_MAX is overkill
// but it's also the *correct* thing to do.
int low_condition = (cdf(low) < p);
int high_condition = (p < cdf(high));
if (low_condition && high_condition) {
interval_found = 1;
} else if (!low_condition) {
low = low * 2;
} else if (!high_condition) {
high = high * 2;
}
}
if (!interval_found) {
PROCESS_ERROR("Interval containing the target value not found, in function inverse_cdf");
} else {
int convergence_condition = 0;
int count = 0;
while (!convergence_condition && (count < (INT_MAX / 2))) {
float mid = (high + low) / 2;
int mid_not_new = (mid == low) || (mid == high);
// float width = high - low;
// if ((width < 1e-8) || mid_not_new){
if (mid_not_new) {
convergence_condition = 1;
} else {
float mid_sign = cdf(mid) - p;
if (mid_sign < 0) {
low = mid;
} else if (mid_sign > 0) {
high = mid;
} else if (mid_sign == 0) {
low = mid;
high = mid;
}
}
}
if (convergence_condition) {
struct box result = { .empty = 0, .content = low };
return result;
} else {
PROCESS_ERROR("Search process did not converge, in function inverse_cdf");
}
}
}
struct box inverse_cdf_box(struct box cdf_box(float), float p)
{
// given a cdf: [-Inf, Inf] => Box([0,1])
// returns a box with either
// x such that cdf(x) = p
// or an error
// if EXIT_ON_ERROR is set to 1, it exits instead of providing an error
float low = -1.0;
float high = 1.0;
// 1. Make sure that cdf(low) < p < cdf(high)
int interval_found = 0;
while ((!interval_found) && (low > -FLT_MAX / 4) && (high < FLT_MAX / 4)) {
// ^ Using FLT_MIN and FLT_MAX is overkill
// but it's also the *correct* thing to do.
struct box cdf_low = cdf_box(low);
if (cdf_low.empty) {
PROCESS_ERROR(cdf_low.error_msg);
}
struct box cdf_high = cdf_box(high);
if (cdf_high.empty) {
PROCESS_ERROR(cdf_low.error_msg);
}
int low_condition = (cdf_low.content < p);
int high_condition = (p < cdf_high.content);
if (low_condition && high_condition) {
interval_found = 1;
} else if (!low_condition) {
low = low * 2;
} else if (!high_condition) {
high = high * 2;
}
}
if (!interval_found) {
PROCESS_ERROR("Interval containing the target value not found, in function inverse_cdf");
} else {
int convergence_condition = 0;
int count = 0;
while (!convergence_condition && (count < (INT_MAX / 2))) {
float mid = (high + low) / 2;
int mid_not_new = (mid == low) || (mid == high);
// float width = high - low;
if (mid_not_new) {
// if ((width < 1e-8) || mid_not_new){
convergence_condition = 1;
} else {
struct box cdf_mid = cdf_box(mid);
if (cdf_mid.empty) {
PROCESS_ERROR(cdf_mid.error_msg);
}
float mid_sign = cdf_mid.content - p;
if (mid_sign < 0) {
low = mid;
} else if (mid_sign > 0) {
high = mid;
} else if (mid_sign == 0) {
low = mid;
high = mid;
}
}
}
if (convergence_condition) {
struct box result = { .empty = 0, .content = low };
return result;
} else {
PROCESS_ERROR("Search process did not converge, in function inverse_cdf");
}
}
}
// Some randomness functions for:
// - Sampling from a cdf
// - Benchmarking against a previous approach, which will be faster, but less general
// Get random number between 0 and 1
uint32_t xorshift32(uint32_t* seed)
{
// Algorithm "xor" from p. 4 of Marsaglia, "Xorshift RNGs"
// See <https://stackoverflow.com/questions/53886131/how-does-xorshift32-works>
// https://en.wikipedia.org/wiki/Xorshift
// Also some drama: <https://www.pcg-random.org/posts/on-vignas-pcg-critique.html>, <https://prng.di.unimi.it/>
uint32_t x = *seed;
x ^= x << 13;
x ^= x >> 17;
x ^= x << 5;
return *seed = x;
}
// Distribution & sampling functions
float rand_0_to_1(uint32_t* seed)
{
return ((float)xorshift32(seed)) / ((float)UINT32_MAX);
}
// Sampler based on inverse cdf and randomness function
struct box sampler_box_cdf(struct box cdf(float), uint32_t* seed)
{
float p = rand_0_to_1(seed);
struct box result = inverse_cdf_box(cdf, p);
return result;
}
struct box sampler_float_cdf(float cdf(float), uint32_t* seed)
{
float p = rand_0_to_1(seed);
struct box result = inverse_cdf_float(cdf, p);
return result;
}
// Comparison point with raw normal sampler
const float PI = 3.14159265358979323846;
float sampler_normal_0_1(uint32_t* seed)
{
float u1 = rand_0_to_1(seed);
float u2 = rand_0_to_1(seed);
float z = sqrtf(-2.0 * log(u1)) * sin(2 * PI * u2);
return z;
}
// Some testers // Some testers
void test_inverse_cdf_float(char* cdf_name, float cdf_float(float)) void test_inverse_cdf_float(char* cdf_name, float cdf_float(float))
{ {
@ -445,12 +225,12 @@ int main()
test_and_time_sampler_float("cdf_normal_0_1", cdf_normal_0_1, seed); test_and_time_sampler_float("cdf_normal_0_1", cdf_normal_0_1, seed);
// Get some normal samples using a previous approach // Get some normal samples using a previous approach
printf("\nGetting some samples from sampler_normal_0_1\n"); printf("\nGetting some samples from unit_normal\n");
clock_t begin_2 = clock(); clock_t begin_2 = clock();
for (int i = 0; i < NUM_SAMPLES; i++) { for (int i = 0; i < NUM_SAMPLES; i++) {
float normal_sample = sampler_normal_0_1(seed); float normal_sample = unit_normal(seed);
// printf("%f\n", normal_sample); // printf("%f\n", normal_sample);
} }
@ -460,11 +240,11 @@ int main()
// Test box sampler // Test box sampler
test_and_time_sampler_box("cdf_beta", cdf_beta, seed); test_and_time_sampler_box("cdf_beta", cdf_beta, seed);
// Ok, this is slower than python!! // Ok, this is slower than python!!
// Partly this is because I am using a more general algorithm, // Partly this is because I am using a more general algorithm,
// which applies to any cdf // which applies to any cdf
// But I am also using really anal convergence conditions. // But I am also using really anal convergence conditions.
// This could be optimized. // This could be optimized.
free(seed); free(seed);
return 0; return 0;

View File

@ -1,6 +1,25 @@
#include <math.h> #include <math.h>
#include <stdint.h> #include <stdint.h>
#include <stdio.h>
#include <stdlib.h> #include <stdlib.h>
#include <float.h>
#include <limits.h>
#include <time.h>
// #include <sys/types.h>
#define EXIT_ON_ERROR 0
#define MAX_ERROR_LENGTH 500
#define PROCESS_ERROR(...) \
do { \
if (EXIT_ON_ERROR) { \
printf("@, in %s (%d)", __FILE__, __LINE__); \
exit(1); \
} else { \
char error_msg[MAX_ERROR_LENGTH]; \
snprintf(error_msg, MAX_ERROR_LENGTH, "@, in %s (%d)", __FILE__, __LINE__); \
struct box error = { .empty = 1, .error_msg = error_msg }; \
return error; \
} \
} while (0)
// PI constant // PI constant
const float PI = M_PI; // 3.14159265358979323846; const float PI = M_PI; // 3.14159265358979323846;
@ -112,3 +131,171 @@ float mixture(float (*samplers[])(uint32_t*), float* weights, int n_dists, uint3
free(cumsummed_normalized_weights); free(cumsummed_normalized_weights);
return result; return result;
} }
// Sample from an arbitrary cdf
struct box {
int empty;
float content;
char* error_msg;
};
// Inverse cdf at point
// Two versions of this function:
// - raw, dealing with cdfs that return floats
// - input: cdf: float => float, p
// - output: Box(number|error)
// - box, dealing with cdfs that return a box.
// - input: cdf: float => Box(number|error), p
// - output: Box(number|error)
struct box inverse_cdf_float(float cdf(float), float p)
{
// given a cdf: [-Inf, Inf] => [0,1]
// returns a box with either
// x such that cdf(x) = p
// or an error
// if EXIT_ON_ERROR is set to 1, it exits instead of providing an error
float low = -1.0;
float high = 1.0;
// 1. Make sure that cdf(low) < p < cdf(high)
int interval_found = 0;
while ((!interval_found) && (low > -FLT_MAX / 4) && (high < FLT_MAX / 4)) {
// ^ Using FLT_MIN and FLT_MAX is overkill
// but it's also the *correct* thing to do.
int low_condition = (cdf(low) < p);
int high_condition = (p < cdf(high));
if (low_condition && high_condition) {
interval_found = 1;
} else if (!low_condition) {
low = low * 2;
} else if (!high_condition) {
high = high * 2;
}
}
if (!interval_found) {
PROCESS_ERROR("Interval containing the target value not found, in function inverse_cdf");
} else {
int convergence_condition = 0;
int count = 0;
while (!convergence_condition && (count < (INT_MAX / 2))) {
float mid = (high + low) / 2;
int mid_not_new = (mid == low) || (mid == high);
// float width = high - low;
// if ((width < 1e-8) || mid_not_new){
if (mid_not_new) {
convergence_condition = 1;
} else {
float mid_sign = cdf(mid) - p;
if (mid_sign < 0) {
low = mid;
} else if (mid_sign > 0) {
high = mid;
} else if (mid_sign == 0) {
low = mid;
high = mid;
}
}
}
if (convergence_condition) {
struct box result = { .empty = 0, .content = low };
return result;
} else {
PROCESS_ERROR("Search process did not converge, in function inverse_cdf");
}
}
}
struct box inverse_cdf_box(struct box cdf_box(float), float p)
{
// given a cdf: [-Inf, Inf] => Box([0,1])
// returns a box with either
// x such that cdf(x) = p
// or an error
// if EXIT_ON_ERROR is set to 1, it exits instead of providing an error
float low = -1.0;
float high = 1.0;
// 1. Make sure that cdf(low) < p < cdf(high)
int interval_found = 0;
while ((!interval_found) && (low > -FLT_MAX / 4) && (high < FLT_MAX / 4)) {
// ^ Using FLT_MIN and FLT_MAX is overkill
// but it's also the *correct* thing to do.
struct box cdf_low = cdf_box(low);
if (cdf_low.empty) {
PROCESS_ERROR(cdf_low.error_msg);
}
struct box cdf_high = cdf_box(high);
if (cdf_high.empty) {
PROCESS_ERROR(cdf_low.error_msg);
}
int low_condition = (cdf_low.content < p);
int high_condition = (p < cdf_high.content);
if (low_condition && high_condition) {
interval_found = 1;
} else if (!low_condition) {
low = low * 2;
} else if (!high_condition) {
high = high * 2;
}
}
if (!interval_found) {
PROCESS_ERROR("Interval containing the target value not found, in function inverse_cdf");
} else {
int convergence_condition = 0;
int count = 0;
while (!convergence_condition && (count < (INT_MAX / 2))) {
float mid = (high + low) / 2;
int mid_not_new = (mid == low) || (mid == high);
// float width = high - low;
if (mid_not_new) {
// if ((width < 1e-8) || mid_not_new){
convergence_condition = 1;
} else {
struct box cdf_mid = cdf_box(mid);
if (cdf_mid.empty) {
PROCESS_ERROR(cdf_mid.error_msg);
}
float mid_sign = cdf_mid.content - p;
if (mid_sign < 0) {
low = mid;
} else if (mid_sign > 0) {
high = mid;
} else if (mid_sign == 0) {
low = mid;
high = mid;
}
}
}
if (convergence_condition) {
struct box result = { .empty = 0, .content = low };
return result;
} else {
PROCESS_ERROR("Search process did not converge, in function inverse_cdf");
}
}
}
// Sampler based on inverse cdf and randomness function
struct box sampler_box_cdf(struct box cdf(float), uint32_t* seed)
{
float p = rand_0_to_1(seed);
struct box result = inverse_cdf_box(cdf, p);
return result;
}
struct box sampler_float_cdf(float cdf(float), uint32_t* seed)
{
float p = rand_0_to_1(seed);
struct box result = inverse_cdf_float(cdf, p);
return result;
}

View File

@ -4,13 +4,29 @@
// uint32_t header // uint32_t header
#include <stdint.h> #include <stdint.h>
// Macros
#define EXIT_ON_ERROR 0
#define MAX_ERROR_LENGTH 500
#define PROCESS_ERROR(...) \
do { \
if (EXIT_ON_ERROR) { \
printf("@, in %s (%d)", __FILE__, __LINE__); \
exit(1); \
} else { \
char error_msg[MAX_ERROR_LENGTH]; \
snprintf(error_msg, MAX_ERROR_LENGTH, "@, in %s (%d)", __FILE__, __LINE__); \
struct box error = { .empty = 1, .error_msg = error_msg }; \
return error; \
} \
} while (0)
// Pseudo Random number generator // Pseudo Random number generator
uint32_t xorshift32(uint32_t* seed); uint32_t xorshift32(uint32_t* seed);
// Distribution & sampling functions // Distribution & sampling functions
float rand_0_to_1(uint32_t* seed); float rand_0_to_1(uint32_t* seed);
float rand_float(float max, uint32_t* seed); float rand_float(float max, uint32_t* seed);
float ur_normal(uint32_t* seed); float unit_normal(uint32_t* seed);
float random_uniform(float from, float to, uint32_t* seed); float random_uniform(float from, float to, uint32_t* seed);
float random_normal(float mean, float sigma, uint32_t* seed); float random_normal(float mean, float sigma, uint32_t* seed);
float random_lognormal(float logmean, float logsigma, uint32_t* seed); float random_lognormal(float logmean, float logsigma, uint32_t* seed);
@ -23,4 +39,20 @@ void array_cumsum(float* array_to_sum, float* array_cumsummed, int length);
// Mixture function // Mixture function
float mixture(float (*samplers[])(uint32_t*), float* weights, int n_dists, uint32_t* seed); float mixture(float (*samplers[])(uint32_t*), float* weights, int n_dists, uint32_t* seed);
// Box
struct box {
int empty;
float content;
char* error_msg;
};
// Inverse cdf
struct box inverse_cdf_float(float cdf(float), float p);
struct box inverse_cdf_box(struct box cdf_box(float), float p);
// Samplers from cdf
struct box sampler_box_cdf(struct box cdf(float), uint32_t* seed);
struct box sampler_float_cdf(float cdf(float), uint32_t* seed);
#endif #endif