比较两个双精度浮点数或两个浮点数最有效的方法是什么?
简单地这样做是不正确的:
bool CompareDoubles1 (double A, double B)
{
return A == B;
}
比如:
bool CompareDoubles2 (double A, double B)
{
diff = A - B;
return (diff < EPSILON) && (-diff < EPSILON);
}
似乎是浪费加工。
有人知道更聪明的浮点比较器吗?
在https://en.cppreference.com/w/cpp/types/numeric_limits/epsilon上找到了另一个有趣的实现
#include <cmath>
#include <limits>
#include <iomanip>
#include <iostream>
#include <type_traits>
#include <algorithm>
template<class T>
typename std::enable_if<!std::numeric_limits<T>::is_integer, bool>::type
almost_equal(T x, T y, int ulp)
{
// the machine epsilon has to be scaled to the magnitude of the values used
// and multiplied by the desired precision in ULPs (units in the last place)
return std::fabs(x-y) <= std::numeric_limits<T>::epsilon() * std::fabs(x+y) * ulp
// unless the result is subnormal
|| std::fabs(x-y) < std::numeric_limits<T>::min();
}
int main()
{
double d1 = 0.2;
double d2 = 1 / std::sqrt(5) / std::sqrt(5);
std::cout << std::fixed << std::setprecision(20)
<< "d1=" << d1 << "\nd2=" << d2 << '\n';
if(d1 == d2)
std::cout << "d1 == d2\n";
else
std::cout << "d1 != d2\n";
if(almost_equal(d1, d2, 2))
std::cout << "d1 almost equals d2\n";
else
std::cout << "d1 does not almost equal d2\n";
}
就数量的规模而言:
如果在某种物理意义上,ε是量的大小(即相对值)的一小部分,而A和B类型在同一意义上具有可比性,那么我认为,下面的观点是相当正确的:
#include <limits>
#include <iomanip>
#include <iostream>
#include <cmath>
#include <cstdlib>
#include <cassert>
template< typename A, typename B >
inline
bool close_enough(A const & a, B const & b,
typename std::common_type< A, B >::type const & epsilon)
{
using std::isless;
assert(isless(0, epsilon)); // epsilon is a part of the whole quantity
assert(isless(epsilon, 1));
using std::abs;
auto const delta = abs(a - b);
auto const x = abs(a);
auto const y = abs(b);
// comparable generally and |a - b| < eps * (|a| + |b|) / 2
return isless(epsilon * y, x) && isless(epsilon * x, y) && isless((delta + delta) / (x + y), epsilon);
}
int main()
{
std::cout << std::boolalpha << close_enough(0.9, 1.0, 0.1) << std::endl;
std::cout << std::boolalpha << close_enough(1.0, 1.1, 0.1) << std::endl;
std::cout << std::boolalpha << close_enough(1.1, 1.2, 0.01) << std::endl;
std::cout << std::boolalpha << close_enough(1.0001, 1.0002, 0.01) << std::endl;
std::cout << std::boolalpha << close_enough(1.0, 0.01, 0.1) << std::endl;
return EXIT_SUCCESS;
}
我的课程是基于之前发布的答案。非常类似于谷歌的代码,但我使用了一个偏差,将所有NaN值推到0xFF000000以上。这样可以更快地检查NaN。
这段代码是为了演示概念,而不是通用的解决方案。谷歌的代码已经展示了如何计算所有平台特定的值,我不想复制所有这些。我对这段代码做了有限的测试。
typedef unsigned int U32;
// Float Memory Bias (unsigned)
// ----- ------ ---------------
// NaN 0xFFFFFFFF 0xFF800001
// NaN 0xFF800001 0xFFFFFFFF
// -Infinity 0xFF800000 0x00000000 ---
// -3.40282e+038 0xFF7FFFFF 0x00000001 |
// -1.40130e-045 0x80000001 0x7F7FFFFF |
// -0.0 0x80000000 0x7F800000 |--- Valid <= 0xFF000000.
// 0.0 0x00000000 0x7F800000 | NaN > 0xFF000000
// 1.40130e-045 0x00000001 0x7F800001 |
// 3.40282e+038 0x7F7FFFFF 0xFEFFFFFF |
// Infinity 0x7F800000 0xFF000000 ---
// NaN 0x7F800001 0xFF000001
// NaN 0x7FFFFFFF 0xFF7FFFFF
//
// Either value of NaN returns false.
// -Infinity and +Infinity are not "close".
// -0 and +0 are equal.
//
class CompareFloat{
public:
union{
float m_f32;
U32 m_u32;
};
static bool CompareFloat::IsClose( float A, float B, U32 unitsDelta = 4 )
{
U32 a = CompareFloat::GetBiased( A );
U32 b = CompareFloat::GetBiased( B );
if ( (a > 0xFF000000) || (b > 0xFF000000) )
{
return( false );
}
return( (static_cast<U32>(abs( a - b ))) < unitsDelta );
}
protected:
static U32 CompareFloat::GetBiased( float f )
{
U32 r = ((CompareFloat*)&f)->m_u32;
if ( r & 0x80000000 )
{
return( ~r - 0x007FFFFF );
}
return( r + 0x7F800000 );
}
};
我最终花了相当多的时间在这个伟大的线程通过材料。我怀疑每个人都想花这么多时间,所以我将强调我所学到的总结和我实施的解决方案。
快速的总结
Is 1e-8 approximately same as 1e-16? If you are looking at noisy sensor data then probably yes but if you are doing molecular simulation then may be not! Bottom line: You always need to think of tolerance value in context of specific function call and not just make it generic app-wide hard-coded constant.
For general library functions, it's still nice to have parameter with default tolerance. A typical choice is numeric_limits::epsilon() which is same as FLT_EPSILON in float.h. This is however problematic because epsilon for comparing values like 1.0 is not same as epsilon for values like 1E9. The FLT_EPSILON is defined for 1.0.
The obvious implementation to check if number is within tolerance is fabs(a-b) <= epsilon however this doesn't work because default epsilon is defined for 1.0. We need to scale epsilon up or down in terms of a and b.
There are two solution to this problem: either you set epsilon proportional to max(a,b) or you can get next representable numbers around a and then see if b falls into that range. The former is called "relative" method and later is called ULP method.
Both methods actually fails anyway when comparing with 0. In this case, application must supply correct tolerance.
实用函数实现(c++ 11)
//implements relative method - do not use for comparing with zero
//use this most of the time, tolerance needs to be meaningful in your context
template<typename TReal>
static bool isApproximatelyEqual(TReal a, TReal b, TReal tolerance = std::numeric_limits<TReal>::epsilon())
{
TReal diff = std::fabs(a - b);
if (diff <= tolerance)
return true;
if (diff < std::fmax(std::fabs(a), std::fabs(b)) * tolerance)
return true;
return false;
}
//supply tolerance that is meaningful in your context
//for example, default tolerance may not work if you are comparing double with float
template<typename TReal>
static bool isApproximatelyZero(TReal a, TReal tolerance = std::numeric_limits<TReal>::epsilon())
{
if (std::fabs(a) <= tolerance)
return true;
return false;
}
//use this when you want to be on safe side
//for example, don't start rover unless signal is above 1
template<typename TReal>
static bool isDefinitelyLessThan(TReal a, TReal b, TReal tolerance = std::numeric_limits<TReal>::epsilon())
{
TReal diff = a - b;
if (diff < tolerance)
return true;
if (diff < std::fmax(std::fabs(a), std::fabs(b)) * tolerance)
return true;
return false;
}
template<typename TReal>
static bool isDefinitelyGreaterThan(TReal a, TReal b, TReal tolerance = std::numeric_limits<TReal>::epsilon())
{
TReal diff = a - b;
if (diff > tolerance)
return true;
if (diff > std::fmax(std::fabs(a), std::fabs(b)) * tolerance)
return true;
return false;
}
//implements ULP method
//use this when you are only concerned about floating point precision issue
//for example, if you want to see if a is 1.0 by checking if its within
//10 closest representable floating point numbers around 1.0.
template<typename TReal>
static bool isWithinPrecisionInterval(TReal a, TReal b, unsigned int interval_size = 1)
{
TReal min_a = a - (a - std::nextafter(a, std::numeric_limits<TReal>::lowest())) * interval_size;
TReal max_a = a + (std::nextafter(a, std::numeric_limits<TReal>::max()) - a) * interval_size;
return min_a <= b && max_a >= b;
}
下面是使用std::numeric_limits::epsilon()不是答案的证明——对于大于1的值它会失败:
证明我上面的评论:
#include <stdio.h>
#include <limits>
double ItoD (__int64 x) {
// Return double from 64-bit hexadecimal representation.
return *(reinterpret_cast<double*>(&x));
}
void test (__int64 ai, __int64 bi) {
double a = ItoD(ai), b = ItoD(bi);
bool close = std::fabs(a-b) < std::numeric_limits<double>::epsilon();
printf ("%.16f and %.16f %s close.\n", a, b, close ? "are " : "are not");
}
int main()
{
test (0x3fe0000000000000L,
0x3fe0000000000001L);
test (0x3ff0000000000000L,
0x3ff0000000000001L);
}
运行产生如下输出:
0.5000000000000000 and 0.5000000000000001 are close.
1.0000000000000000 and 1.0000000000000002 are not close.
请注意,在第二种情况下(1且仅大于1),两个输入值尽可能接近,并且仍然比较为不接近。因此,对于大于1.0的值,不妨只使用相等性测试。当比较浮点值时,固定的epsilon将无法保存您的数据。