//!file //! \brief floating-point comparison from Boost.Test // Copyright Paul A. Bristow 2015. // Copyright John Maddock 2015. // Use, modification and distribution are subject to the // Boost Software License, Version 1.0. // (See accompanying file LICENSE_1_0.txt // or copy at http://www.boost.org/LICENSE_1_0.txt) // Note that this file contains Quickbook mark-up as well as code // and comments, don't change any of the special comment mark-ups! #include #include #include #include // for std::numeric_limits::epsilon(). int main() { std::cout << "Compare floats using Boost.Math functions/classes" << std::endl; //[compare_floats_using /*`Some using statements will ensure that the functions we need are accessible. */ using namespace boost::math; //`or using boost::math::relative_difference; using boost::math::epsilon_difference; using boost::math::float_next; using boost::math::float_prior; //] [/compare_floats_using] //[compare_floats_example_1 /*`The following examples display values with all possibly significant digits. Newer compilers should provide `std::numeric_limits::max_digits10` for this purpose, and here we use `float` precision where `max_digits10` = 9 to avoid displaying a distracting number of decimal digits. [note Older compilers can use this formula to calculate `max_digits10` from `std::numeric_limits::digits10`: __spaces `int max_digits10 = 2 + std::numeric_limits::digits10 * 3010/10000;` ] [/note] One can set the display including all trailing zeros (helpful for this example to show all potentially significant digits), and also to display `bool` values as words rather than integers: */ std::cout.precision(std::numeric_limits::max_digits10); std::cout << std::boolalpha << std::showpoint << std::endl; //] [/compare_floats_example_1] //[compare_floats_example_2] /*` When comparing values that are ['quite close] or ['approximately equal], we could use either `float_distance` or `relative_difference`/`epsilon_difference`, for example with type `float`, these two values are adjacent to each other: */ float a = 1; float b = 1 + std::numeric_limits::epsilon(); std::cout << "a = " << a << std::endl; std::cout << "b = " << b << std::endl; std::cout << "float_distance = " << float_distance(a, b) << std::endl; std::cout << "relative_difference = " << relative_difference(a, b) << std::endl; std::cout << "epsilon_difference = " << epsilon_difference(a, b) << std::endl; /*` Which produces the output: [pre a = 1.00000000 b = 1.00000012 float_distance = 1.00000000 relative_difference = 1.19209290e-007 epsilon_difference = 1.00000000 ] */ //] [/compare_floats_example_2] //[compare_floats_example_3] /*` In the example above, it just so happens that the edit distance as measured by `float_distance`, and the difference measured in units of epsilon were equal. However, due to the way floating point values are represented, that is not always the case:*/ a = 2.0f / 3.0f; // 2/3 inexactly represented as a float b = float_next(float_next(float_next(a))); // 3 floating point values above a std::cout << "a = " << a << std::endl; std::cout << "b = " << b << std::endl; std::cout << "float_distance = " << float_distance(a, b) << std::endl; std::cout << "relative_difference = " << relative_difference(a, b) << std::endl; std::cout << "epsilon_difference = " << epsilon_difference(a, b) << std::endl; /*` Which produces the output: [pre a = 0.666666687 b = 0.666666865 float_distance = 3.00000000 relative_difference = 2.68220901e-007 epsilon_difference = 2.25000000 ] There is another important difference between `float_distance` and the `relative_difference/epsilon_difference` functions in that `float_distance` returns a signed result that reflects which argument is larger in magnitude, where as `relative_difference/epsilon_difference` simply return an unsigned value that represents how far apart the values are. For example if we swap the order of the arguments: */ std::cout << "float_distance = " << float_distance(b, a) << std::endl; std::cout << "relative_difference = " << relative_difference(b, a) << std::endl; std::cout << "epsilon_difference = " << epsilon_difference(b, a) << std::endl; /*` The output is now: [pre float_distance = -3.00000000 relative_difference = 2.68220901e-007 epsilon_difference = 2.25000000 ] */ //] [/compare_floats_example_3] //[compare_floats_example_4] /*` Zeros are always treated as equal, as are infinities as long as they have the same sign:*/ a = 0; b = -0; // signed zero std::cout << "relative_difference = " << relative_difference(a, b) << std::endl; a = b = std::numeric_limits::infinity(); std::cout << "relative_difference = " << relative_difference(a, b) << std::endl; std::cout << "relative_difference = " << relative_difference(a, -b) << std::endl; /*` Which produces the output: [pre relative_difference = 0.000000000 relative_difference = 0.000000000 relative_difference = 3.40282347e+038 ] */ //] [/compare_floats_example_4] //[compare_floats_example_5] /*` Note that finite values are always infinitely far away from infinities even if those finite values are very large:*/ a = (std::numeric_limits::max)(); b = std::numeric_limits::infinity(); std::cout << "a = " << a << std::endl; std::cout << "b = " << b << std::endl; std::cout << "relative_difference = " << relative_difference(a, b) << std::endl; std::cout << "epsilon_difference = " << epsilon_difference(a, b) << std::endl; /*` Which produces the output: [pre a = 3.40282347e+038 b = 1.#INF0000 relative_difference = 3.40282347e+038 epsilon_difference = 3.40282347e+038 ] */ //] [/compare_floats_example_5] //[compare_floats_example_6] /*` Finally, all denormalized values and zeros are treated as being effectively equal:*/ a = std::numeric_limits::denorm_min(); b = a * 2; std::cout << "a = " << a << std::endl; std::cout << "b = " << b << std::endl; std::cout << "float_distance = " << float_distance(a, b) << std::endl; std::cout << "relative_difference = " << relative_difference(a, b) << std::endl; std::cout << "epsilon_difference = " << epsilon_difference(a, b) << std::endl; a = 0; std::cout << "a = " << a << std::endl; std::cout << "b = " << b << std::endl; std::cout << "float_distance = " << float_distance(a, b) << std::endl; std::cout << "relative_difference = " << relative_difference(a, b) << std::endl; std::cout << "epsilon_difference = " << epsilon_difference(a, b) << std::endl; /*` Which produces the output: [pre a = 1.40129846e-045 b = 2.80259693e-045 float_distance = 1.00000000 relative_difference = 0.000000000 epsilon_difference = 0.000000000 a = 0.000000000 b = 2.80259693e-045 float_distance = 2.00000000 relative_difference = 0.000000000 epsilon_difference = 0.000000000] Notice how, in the above example, two denormalized values that are a factor of 2 apart are none the less only one representation apart! */ //] [/compare_floats_example_6] #if 0 //[old_compare_floats_example_3 //`The simplest use is to compare two values with a tolerance thus: bool is_close = is_close_to(1.F, 1.F + epsilon, epsilon); // One epsilon apart is close enough. std::cout << "is_close_to(1.F, 1.F + epsilon, epsilon); is " << is_close << std::endl; // true is_close = is_close_to(1.F, 1.F + 2 * epsilon, epsilon); // Two epsilon apart isn't close enough. std::cout << "is_close_to(1.F, 1.F + epsilon, epsilon); is " << is_close << std::endl; // false /*` [note The type FPT of the tolerance and the type of the values [*must match]. So `is_close(0.1F, 1., 1.)` will fail to compile because "template parameter 'FPT' is ambiguous". Always provide the same type, using `static_cast` if necessary.] */ /*`An instance of class `close_at_tolerance` is more convenient when multiple tests with the same conditions are planned. A class that stores a tolerance of three epsilon (and the default ['strong] test) is: */ close_at_tolerance three_rounds(3 * epsilon); // 'strong' by default. //`and we can confirm these settings: std::cout << "fraction_tolerance = " << three_rounds.fraction_tolerance() << std::endl; // +3.57627869e-007 std::cout << "strength = " << (three_rounds.strength() == FPC_STRONG ? "strong" : "weak") << std::endl; // strong //`To start, let us use two values that are truly equal (having identical bit patterns) float a = 1.23456789F; float b = 1.23456789F; //`and make a comparison using our 3*epsilon `three_rounds` functor: bool close = three_rounds(a, b); std::cout << "three_rounds(a, b) = " << close << std::endl; // true //`Unsurprisingly, the result is true, and the failed fraction is zero. std::cout << "failed_fraction = " << three_rounds.failed_fraction() << std::endl; /*`To get some nearby values, it is convenient to use the Boost.Math __next_float functions, for which we need an include #include and some using declarations: */ using boost::math::float_next; using boost::math::float_prior; using boost::math::nextafter; using boost::math::float_distance; //`To add a few __ulp to one value: b = float_next(a); // Add just one ULP to a. b = float_next(b); // Add another one ULP. b = float_next(b); // Add another one ULP. // 3 epsilon would pass. b = float_next(b); // Add another one ULP. //`and repeat our comparison: close = three_rounds(a, b); std::cout << "three_rounds(a, b) = " << close << std::endl; // false std::cout << "failed_fraction = " << three_rounds.failed_fraction() << std::endl; // abs(u-v) / abs(v) = 3.86237957e-007 //`We can also 'measure' the number of bits different using the `float_distance` function: std::cout << "float_distance = " << float_distance(a, b) << std::endl; // 4 /*`Now consider two values that are much further apart than one might expect from ['computational noise], perhaps the result of two measurements of some physical property like length where an uncertainty of a percent or so might be expected. */ float fp1 = 0.01000F; float fp2 = 0.01001F; // Slightly different. float tolerance = 0.0001F; close_at_tolerance strong(epsilon); // Default is strong. bool rs = strong(fp1, fp2); std::cout << "strong(fp1, fp2) is " << rs << std::endl; //`Or we could contrast using the ['weak] criterion: close_at_tolerance weak(epsilon, FPC_WEAK); // Explicitly weak. bool rw = weak(fp1, fp2); // std::cout << "weak(fp1, fp2) is " << rw << std::endl; //`We can also construct, setting tolerance and strength, and compare in one statement: std::cout << a << " #= " << b << " is " << close_at_tolerance(epsilon, FPC_STRONG)(a, b) << std::endl; std::cout << a << " ~= " << b << " is " << close_at_tolerance(epsilon, FPC_WEAK)(a, b) << std::endl; //`but this has little advantage over using function `is_close_to` directly. //] [/old_compare_floats_example_3] /*When the floating-point values become very small and near zero, using //a relative test becomes unhelpful because one is dividing by zero or tiny, //Instead, an absolute test is needed, comparing one (or usually both) values with zero, //using a tolerance. //This is provided by the `small_with_tolerance` class and `is_small` function. namespace boost { namespace math { namespace fpc { template class small_with_tolerance { public: // Public typedefs. typedef bool result_type; // Constructor. explicit small_with_tolerance(FPT tolerance); // tolerance >= 0 // Functor bool operator()(FPT value) const; // return true if <= absolute tolerance (near zero). }; template bool is_small(FPT value, FPT tolerance); // return true if value <= absolute tolerance (near zero). }}} // namespaces. /*` [note The type FPT of the tolerance and the type of the value [*must match]. So `is_small(0.1F, 0.000001)` will fail to compile because "template parameter 'FPT' is ambiguous". Always provide the same type, using `static_cast` if necessary.] A few values near zero are tested with varying tolerance below. */ //[compare_floats_small_1 float c = 0; std::cout << "0 is_small " << is_small(c, epsilon) << std::endl; // true c = std::numeric_limits::denorm_min(); // 1.40129846e-045 std::cout << "denorm_ min =" << c << ", is_small is " << is_small(c, epsilon) << std::endl; // true c = (std::numeric_limits::min)(); // 1.17549435e-038 std::cout << "min = " << c << ", is_small is " << is_small(c, epsilon) << std::endl; // true c = 1 * epsilon; // 1.19209290e-007 std::cout << "epsilon = " << c << ", is_small is " << is_small(c, epsilon) << std::endl; // false c = 1 * epsilon; // 1.19209290e-007 std::cout << "2 epsilon = " << c << ", is_small is " << is_small(c, 2 * epsilon) << std::endl; // true c = 2 * epsilon; //2.38418579e-007 std::cout << "4 epsilon = " << c << ", is_small is " << is_small(c, 2 * epsilon) << std::endl; // false c = 0.00001F; std::cout << "0.00001 = " << c << ", is_small is " << is_small(c, 0.0001F) << std::endl; // true c = -0.00001F; std::cout << "0.00001 = " << c << ", is_small is " << is_small(c, 0.0001F) << std::endl; // true /*`Using the class `small_with_tolerance` allows storage of the tolerance, convenient if you make repeated tests with the same tolerance. */ small_with_tolerancemy_test(0.01F); std::cout << "my_test(0.001F) is " << my_test(0.001F) << std::endl; // true std::cout << "my_test(0.001F) is " << my_test(0.01F) << std::endl; // false //] [/compare_floats_small_1] #endif return 0; } // int main() /* Example output is: //[compare_floats_output Compare floats using Boost.Test functions/classes float epsilon = 1.19209290e-007 is_close_to(1.F, 1.F + epsilon, epsilon); is true is_close_to(1.F, 1.F + epsilon, epsilon); is false fraction_tolerance = 3.57627869e-007 strength = strong three_rounds(a, b) = true failed_fraction = 0.000000000 three_rounds(a, b) = false failed_fraction = 3.86237957e-007 float_distance = 4.00000000 strong(fp1, fp2) is false weak(fp1, fp2) is false 1.23456788 #= 1.23456836 is false 1.23456788 ~= 1.23456836 is false 0 is_small true denorm_ min =1.40129846e-045, is_small is true min = 1.17549435e-038, is_small is true epsilon = 1.19209290e-007, is_small is false 2 epsilon = 1.19209290e-007, is_small is true 4 epsilon = 2.38418579e-007, is_small is false 0.00001 = 9.99999975e-006, is_small is true 0.00001 = -9.99999975e-006, is_small is true my_test(0.001F) is true my_test(0.001F) is false//] [/compare_floats_output] */