Logistic regression 2

Percentage Accurate: 99.3% → 99.4%

Time: 11.9s

Alternatives: 9

Speedup: 1.8×

• could not determine a ground truth

  1. x = 1.365345049647716e+276
  2. y = 8.041032435427806e-260

Specification

Math

\[\begin{array}{l} \\ \log \left(1 + e^{x}\right) - x \cdot y \end{array} \]

FPCore

(FPCore (x y) :precision binary64 (- (log (+ 1.0 (exp x))) (* x y)))

C

double code(double x, double y) {
	return log((1.0 + exp(x))) - (x * y);
}

Fortran

real(8) function code(x, y)
    real(8), intent (in) :: x
    real(8), intent (in) :: y
    code = log((1.0d0 + exp(x))) - (x * y)
end function

Java

public static double code(double x, double y) {
	return Math.log((1.0 + Math.exp(x))) - (x * y);
}

Python

def code(x, y):
	return math.log((1.0 + math.exp(x))) - (x * y)

Julia

function code(x, y)
	return Float64(log(Float64(1.0 + exp(x))) - Float64(x * y))
end

MATLAB

function tmp = code(x, y)
	tmp = log((1.0 + exp(x))) - (x * y);
end

Wolfram

code[x_, y_] := N[(N[Log[N[(1.0 + N[Exp[x], $MachinePrecision]), $MachinePrecision]], $MachinePrecision] - N[(x * y), $MachinePrecision]), $MachinePrecision]

Initial Program: 99.3% accurate, 1.0× speedup

Math

\[\begin{array}{l} \\ \log \left(1 + e^{x}\right) - x \cdot y \end{array} \]

FPCore

(FPCore (x y) :precision binary64 (- (log (+ 1.0 (exp x))) (* x y)))

C

double code(double x, double y) {
	return log((1.0 + exp(x))) - (x * y);
}

Fortran

real(8) function code(x, y)
    real(8), intent (in) :: x
    real(8), intent (in) :: y
    code = log((1.0d0 + exp(x))) - (x * y)
end function

Java

public static double code(double x, double y) {
	return Math.log((1.0 + Math.exp(x))) - (x * y);
}

Python

def code(x, y):
	return math.log((1.0 + math.exp(x))) - (x * y)

Julia

function code(x, y)
	return Float64(log(Float64(1.0 + exp(x))) - Float64(x * y))
end

MATLAB

function tmp = code(x, y)
	tmp = log((1.0 + exp(x))) - (x * y);
end

Wolfram

code[x_, y_] := N[(N[Log[N[(1.0 + N[Exp[x], $MachinePrecision]), $MachinePrecision]], $MachinePrecision] - N[(x * y), $MachinePrecision]), $MachinePrecision]

Alternative 1: 99.4% accurate, 1.7× speedup

Math

\[\begin{array}{l} \\ \begin{array}{l} \mathbf{if}\;x \leq -2.6:\\ \;\;\;\;0 - y \cdot x\\ \mathbf{else}:\\ \;\;\;\;\log 2 + x \cdot \left(\left(0.5 - y\right) + x \cdot \left(0.125 + -0.005208333333333333 \cdot \left(x \cdot x\right)\right)\right)\\ \end{array} \end{array} \]

FPCore

(FPCore (x y)
 :precision binary64
 (if (<= x -2.6)
   (- 0.0 (* y x))
   (+
    (log 2.0)
    (* x (+ (- 0.5 y) (* x (+ 0.125 (* -0.005208333333333333 (* x x)))))))))

C

double code(double x, double y) {
	double tmp;
	if (x <= -2.6) {
		tmp = 0.0 - (y * x);
	} else {
		tmp = log(2.0) + (x * ((0.5 - y) + (x * (0.125 + (-0.005208333333333333 * (x * x))))));
	}
	return tmp;
}

Fortran

real(8) function code(x, y)
    real(8), intent (in) :: x
    real(8), intent (in) :: y
    real(8) :: tmp
    if (x <= (-2.6d0)) then
        tmp = 0.0d0 - (y * x)
    else
        tmp = log(2.0d0) + (x * ((0.5d0 - y) + (x * (0.125d0 + ((-0.005208333333333333d0) * (x * x))))))
    end if
    code = tmp
end function

Java

public static double code(double x, double y) {
	double tmp;
	if (x <= -2.6) {
		tmp = 0.0 - (y * x);
	} else {
		tmp = Math.log(2.0) + (x * ((0.5 - y) + (x * (0.125 + (-0.005208333333333333 * (x * x))))));
	}
	return tmp;
}

Python

def code(x, y):
	tmp = 0
	if x <= -2.6:
		tmp = 0.0 - (y * x)
	else:
		tmp = math.log(2.0) + (x * ((0.5 - y) + (x * (0.125 + (-0.005208333333333333 * (x * x))))))
	return tmp

Julia

function code(x, y)
	tmp = 0.0
	if (x <= -2.6)
		tmp = Float64(0.0 - Float64(y * x));
	else
		tmp = Float64(log(2.0) + Float64(x * Float64(Float64(0.5 - y) + Float64(x * Float64(0.125 + Float64(-0.005208333333333333 * Float64(x * x)))))));
	end
	return tmp
end

MATLAB

function tmp_2 = code(x, y)
	tmp = 0.0;
	if (x <= -2.6)
		tmp = 0.0 - (y * x);
	else
		tmp = log(2.0) + (x * ((0.5 - y) + (x * (0.125 + (-0.005208333333333333 * (x * x))))));
	end
	tmp_2 = tmp;
end

Wolfram

code[x_, y_] := If[LessEqual[x, -2.6], N[(0.0 - N[(y * x), $MachinePrecision]), $MachinePrecision], N[(N[Log[2.0], $MachinePrecision] + N[(x * N[(N[(0.5 - y), $MachinePrecision] + N[(x * N[(0.125 + N[(-0.005208333333333333 * N[(x * x), $MachinePrecision]), $MachinePrecision]), $MachinePrecision]), $MachinePrecision]), $MachinePrecision]), $MachinePrecision]), $MachinePrecision]]

Derivation

1. Split input into 2 regimes

2. if x < -2.60000000000000009

   1. Initial program 99.4%

      \[\log \left(1 + e^{x}\right) - x \cdot y \]
   2. Add Preprocessing

   3. Taylor expanded in x around inf

      \[\leadsto \color{blue}{-1 \cdot \left(x \cdot y\right)} \]
   4. Step-by-step derivation

      1. mul-1-neg N/A

         \[\leadsto \mathsf{neg}\left(x \cdot y\right) \]

      2. neg-sub0 N/A

         \[\leadsto 0 - \color{blue}{x \cdot y} \]

      3. --lowering--.f64 N/A

         \[\leadsto \mathsf{\_.f64}\left(0, \color{blue}{\left(x \cdot y\right)}\right) \]

      4. *-lowering-*.f64 99.4%

         \[\leadsto \mathsf{\_.f64}\left(0, \mathsf{*.f64}\left(x, \color{blue}{y}\right)\right) \]

   5. Simplified 99.4%

      \[\leadsto \color{blue}{0 - x \cdot y} \]

   7. Applied egg-rr 99.4%

      \[\leadsto \color{blue}{-y \cdot x} \]

   if -2.60000000000000009 < x

   1. Initial program 98.8%

      \[\log \left(1 + e^{x}\right) - x \cdot y \]
   2. Add Preprocessing

   3. Taylor expanded in x around 0

      \[\leadsto \color{blue}{\log 2 + x \cdot \left(\left(\frac{1}{2} + x \cdot \left(\frac{1}{8} + \frac{-1}{192} \cdot {x}^{2}\right)\right) - y\right)} \]
   4. Step-by-step derivation

      1. +-lowering-+.f64 N/A

         \[\leadsto \mathsf{+.f64}\left(\log 2, \color{blue}{\left(x \cdot \left(\left(\frac{1}{2} + x \cdot \left(\frac{1}{8} + \frac{-1}{192} \cdot {x}^{2}\right)\right) - y\right)\right)}\right) \]

      2. log-lowering-log.f64 N/A

         \[\leadsto \mathsf{+.f64}\left(\mathsf{log.f64}\left(2\right), \left(\color{blue}{x} \cdot \left(\left(\frac{1}{2} + x \cdot \left(\frac{1}{8} + \frac{-1}{192} \cdot {x}^{2}\right)\right) - y\right)\right)\right) \]

      3. *-lowering-*.f64 N/A

         \[\leadsto \mathsf{+.f64}\left(\mathsf{log.f64}\left(2\right), \mathsf{*.f64}\left(x, \color{blue}{\left(\left(\frac{1}{2} + x \cdot \left(\frac{1}{8} + \frac{-1}{192} \cdot {x}^{2}\right)\right) - y\right)}\right)\right) \]

      4. +-commutative N/A

         \[\leadsto \mathsf{+.f64}\left(\mathsf{log.f64}\left(2\right), \mathsf{*.f64}\left(x, \left(\left(x \cdot \left(\frac{1}{8} + \frac{-1}{192} \cdot {x}^{2}\right) + \frac{1}{2}\right) - y\right)\right)\right) \]

      5. associate--l+ N/A

         \[\leadsto \mathsf{+.f64}\left(\mathsf{log.f64}\left(2\right), \mathsf{*.f64}\left(x, \left(x \cdot \left(\frac{1}{8} + \frac{-1}{192} \cdot {x}^{2}\right) + \color{blue}{\left(\frac{1}{2} - y\right)}\right)\right)\right) \]

      6. +-commutative N/A

         \[\leadsto \mathsf{+.f64}\left(\mathsf{log.f64}\left(2\right), \mathsf{*.f64}\left(x, \left(\left(\frac{1}{2} - y\right) + \color{blue}{x \cdot \left(\frac{1}{8} + \frac{-1}{192} \cdot {x}^{2}\right)}\right)\right)\right) \]

      7. +-lowering-+.f64 N/A

         \[\leadsto \mathsf{+.f64}\left(\mathsf{log.f64}\left(2\right), \mathsf{*.f64}\left(x, \mathsf{+.f64}\left(\left(\frac{1}{2} - y\right), \color{blue}{\left(x \cdot \left(\frac{1}{8} + \frac{-1}{192} \cdot {x}^{2}\right)\right)}\right)\right)\right) \]

      8. --lowering--.f64 N/A

         \[\leadsto \mathsf{+.f64}\left(\mathsf{log.f64}\left(2\right), \mathsf{*.f64}\left(x, \mathsf{+.f64}\left(\mathsf{\_.f64}\left(\frac{1}{2}, y\right), \left(\color{blue}{x} \cdot \left(\frac{1}{8} + \frac{-1}{192} \cdot {x}^{2}\right)\right)\right)\right)\right) \]

      9. *-lowering-*.f64 N/A

         \[\leadsto \mathsf{+.f64}\left(\mathsf{log.f64}\left(2\right), \mathsf{*.f64}\left(x, \mathsf{+.f64}\left(\mathsf{\_.f64}\left(\frac{1}{2}, y\right), \mathsf{*.f64}\left(x, \color{blue}{\left(\frac{1}{8} + \frac{-1}{192} \cdot {x}^{2}\right)}\right)\right)\right)\right) \]

      10. +-lowering-+.f64 N/A

          \[\leadsto \mathsf{+.f64}\left(\mathsf{log.f64}\left(2\right), \mathsf{*.f64}\left(x, \mathsf{+.f64}\left(\mathsf{\_.f64}\left(\frac{1}{2}, y\right), \mathsf{*.f64}\left(x, \mathsf{+.f64}\left(\frac{1}{8}, \color{blue}{\left(\frac{-1}{192} \cdot {x}^{2}\right)}\right)\right)\right)\right)\right) \]

      11. *-lowering-*.f64 N/A

          \[\leadsto \mathsf{+.f64}\left(\mathsf{log.f64}\left(2\right), \mathsf{*.f64}\left(x, \mathsf{+.f64}\left(\mathsf{\_.f64}\left(\frac{1}{2}, y\right), \mathsf{*.f64}\left(x, \mathsf{+.f64}\left(\frac{1}{8}, \mathsf{*.f64}\left(\frac{-1}{192}, \color{blue}{\left({x}^{2}\right)}\right)\right)\right)\right)\right)\right) \]

      12. unpow2 N/A

          \[\leadsto \mathsf{+.f64}\left(\mathsf{log.f64}\left(2\right), \mathsf{*.f64}\left(x, \mathsf{+.f64}\left(\mathsf{\_.f64}\left(\frac{1}{2}, y\right), \mathsf{*.f64}\left(x, \mathsf{+.f64}\left(\frac{1}{8}, \mathsf{*.f64}\left(\frac{-1}{192}, \left(x \cdot \color{blue}{x}\right)\right)\right)\right)\right)\right)\right) \]

      13. *-lowering-*.f64 99.4%

          \[\leadsto \mathsf{+.f64}\left(\mathsf{log.f64}\left(2\right), \mathsf{*.f64}\left(x, \mathsf{+.f64}\left(\mathsf{\_.f64}\left(\frac{1}{2}, y\right), \mathsf{*.f64}\left(x, \mathsf{+.f64}\left(\frac{1}{8}, \mathsf{*.f64}\left(\frac{-1}{192}, \mathsf{*.f64}\left(x, \color{blue}{x}\right)\right)\right)\right)\right)\right)\right) \]

   5. Simplified 99.4%

      \[\leadsto \color{blue}{\log 2 + x \cdot \left(\left(0.5 - y\right) + x \cdot \left(0.125 + -0.005208333333333333 \cdot \left(x \cdot x\right)\right)\right)} \]
3. Recombined 2 regimes into one program.

4. Final simplification 99.4%

   \[\leadsto \begin{array}{l} \mathbf{if}\;x \leq -2.6:\\ \;\;\;\;0 - y \cdot x\\ \mathbf{else}:\\ \;\;\;\;\log 2 + x \cdot \left(\left(0.5 - y\right) + x \cdot \left(0.125 + -0.005208333333333333 \cdot \left(x \cdot x\right)\right)\right)\\ \end{array} \]
5. Add Preprocessing

Alternative 2: 99.4% accurate, 0.7× speedup

Math

\[\begin{array}{l} \\ \mathsf{fma}\left(0 - y, x, \mathsf{log1p}\left(e^{x}\right)\right) \end{array} \]

FPCore

(FPCore (x y) :precision binary64 (fma (- 0.0 y) x (log1p (exp x))))

C

double code(double x, double y) {
	return fma((0.0 - y), x, log1p(exp(x)));
}

Julia

function code(x, y)
	return fma(Float64(0.0 - y), x, log1p(exp(x)))
end

Wolfram

code[x_, y_] := N[(N[(0.0 - y), $MachinePrecision] * x + N[Log[1 + N[Exp[x], $MachinePrecision]], $MachinePrecision]), $MachinePrecision]

Derivation

1. Initial program 99.0%

   \[\log \left(1 + e^{x}\right) - x \cdot y \]
2. Add Preprocessing
3. Step-by-step derivation

   1. sub-neg N/A

      \[\leadsto \log \left(1 + e^{x}\right) + \color{blue}{\left(\mathsf{neg}\left(x \cdot y\right)\right)} \]

   2. +-commutative N/A

      \[\leadsto \left(\mathsf{neg}\left(x \cdot y\right)\right) + \color{blue}{\log \left(1 + e^{x}\right)} \]

   3. *-commutative N/A

      \[\leadsto \left(\mathsf{neg}\left(y \cdot x\right)\right) + \log \left(\color{blue}{1} + e^{x}\right) \]

   4. distribute-lft-neg-in N/A

      \[\leadsto \left(\mathsf{neg}\left(y\right)\right) \cdot x + \log \color{blue}{\left(1 + e^{x}\right)} \]

   5. fma-define N/A

      \[\leadsto \mathsf{fma}\left(\mathsf{neg}\left(y\right), \color{blue}{x}, \log \left(1 + e^{x}\right)\right) \]

   6. fma-lowering-fma.f64 N/A

      \[\leadsto \mathsf{fma.f64}\left(\left(\mathsf{neg}\left(y\right)\right), \color{blue}{x}, \log \left(1 + e^{x}\right)\right) \]

   7. neg-sub0 N/A

      \[\leadsto \mathsf{fma.f64}\left(\left(0 - y\right), x, \log \left(1 + e^{x}\right)\right) \]

   8. --lowering--.f64 N/A

      \[\leadsto \mathsf{fma.f64}\left(\mathsf{\_.f64}\left(0, y\right), x, \log \left(1 + e^{x}\right)\right) \]

   9. log1p-define N/A

      \[\leadsto \mathsf{fma.f64}\left(\mathsf{\_.f64}\left(0, y\right), x, \left(\mathsf{log1p}\left(e^{x}\right)\right)\right) \]

   10. log1p-lowering-log1p.f64 N/A

       \[\leadsto \mathsf{fma.f64}\left(\mathsf{\_.f64}\left(0, y\right), x, \mathsf{log1p.f64}\left(\left(e^{x}\right)\right)\right) \]

   11. exp-lowering-exp.f64 99.2%

       \[\leadsto \mathsf{fma.f64}\left(\mathsf{\_.f64}\left(0, y\right), x, \mathsf{log1p.f64}\left(\mathsf{exp.f64}\left(x\right)\right)\right) \]

4. Applied egg-rr 99.2%

   \[\leadsto \color{blue}{\mathsf{fma}\left(0 - y, x, \mathsf{log1p}\left(e^{x}\right)\right)} \]
5. Add Preprocessing

Alternative 3: 99.3% accurate, 1.0× speedup

Math

\[\begin{array}{l} \\ \mathsf{log1p}\left(e^{x}\right) - y \cdot x \end{array} \]

FPCore

(FPCore (x y) :precision binary64 (- (log1p (exp x)) (* y x)))

C

double code(double x, double y) {
	return log1p(exp(x)) - (y * x);
}

Java

public static double code(double x, double y) {
	return Math.log1p(Math.exp(x)) - (y * x);
}

Python

def code(x, y):
	return math.log1p(math.exp(x)) - (y * x)

Julia

function code(x, y)
	return Float64(log1p(exp(x)) - Float64(y * x))
end

Wolfram

code[x_, y_] := N[(N[Log[1 + N[Exp[x], $MachinePrecision]], $MachinePrecision] - N[(y * x), $MachinePrecision]), $MachinePrecision]

Derivation

1. Initial program 99.0%

   \[\log \left(1 + e^{x}\right) - x \cdot y \]
2. Step-by-step derivation

   1. --lowering--.f64 N/A

      \[\leadsto \mathsf{\_.f64}\left(\log \left(1 + e^{x}\right), \color{blue}{\left(x \cdot y\right)}\right) \]

   2. log1p-define N/A

      \[\leadsto \mathsf{\_.f64}\left(\left(\mathsf{log1p}\left(e^{x}\right)\right), \left(\color{blue}{x} \cdot y\right)\right) \]

   3. log1p-lowering-log1p.f64 N/A

      \[\leadsto \mathsf{\_.f64}\left(\mathsf{log1p.f64}\left(\left(e^{x}\right)\right), \left(\color{blue}{x} \cdot y\right)\right) \]

   4. exp-lowering-exp.f64 N/A

      \[\leadsto \mathsf{\_.f64}\left(\mathsf{log1p.f64}\left(\mathsf{exp.f64}\left(x\right)\right), \left(x \cdot y\right)\right) \]

   5. *-lowering-*.f64 99.2%

      \[\leadsto \mathsf{\_.f64}\left(\mathsf{log1p.f64}\left(\mathsf{exp.f64}\left(x\right)\right), \mathsf{*.f64}\left(x, \color{blue}{y}\right)\right) \]

3. Simplified 99.2%

   \[\leadsto \color{blue}{\mathsf{log1p}\left(e^{x}\right) - x \cdot y} \]
4. Add Preprocessing

5. Final simplification 99.2%

   \[\leadsto \mathsf{log1p}\left(e^{x}\right) - y \cdot x \]
6. Add Preprocessing

Alternative 4: 99.4% accurate, 1.8× speedup

Math

\[\begin{array}{l} \\ \begin{array}{l} \mathbf{if}\;x \leq -38:\\ \;\;\;\;0 - y \cdot x\\ \mathbf{else}:\\ \;\;\;\;\log 2 + x \cdot \left(0.5 + \left(x \cdot 0.125 - y\right)\right)\\ \end{array} \end{array} \]

FPCore

(FPCore (x y)
 :precision binary64
 (if (<= x -38.0)
   (- 0.0 (* y x))
   (+ (log 2.0) (* x (+ 0.5 (- (* x 0.125) y))))))

C

double code(double x, double y) {
	double tmp;
	if (x <= -38.0) {
		tmp = 0.0 - (y * x);
	} else {
		tmp = log(2.0) + (x * (0.5 + ((x * 0.125) - y)));
	}
	return tmp;
}

Fortran

real(8) function code(x, y)
    real(8), intent (in) :: x
    real(8), intent (in) :: y
    real(8) :: tmp
    if (x <= (-38.0d0)) then
        tmp = 0.0d0 - (y * x)
    else
        tmp = log(2.0d0) + (x * (0.5d0 + ((x * 0.125d0) - y)))
    end if
    code = tmp
end function

Java

public static double code(double x, double y) {
	double tmp;
	if (x <= -38.0) {
		tmp = 0.0 - (y * x);
	} else {
		tmp = Math.log(2.0) + (x * (0.5 + ((x * 0.125) - y)));
	}
	return tmp;
}

Python

def code(x, y):
	tmp = 0
	if x <= -38.0:
		tmp = 0.0 - (y * x)
	else:
		tmp = math.log(2.0) + (x * (0.5 + ((x * 0.125) - y)))
	return tmp

Julia

function code(x, y)
	tmp = 0.0
	if (x <= -38.0)
		tmp = Float64(0.0 - Float64(y * x));
	else
		tmp = Float64(log(2.0) + Float64(x * Float64(0.5 + Float64(Float64(x * 0.125) - y))));
	end
	return tmp
end

MATLAB

function tmp_2 = code(x, y)
	tmp = 0.0;
	if (x <= -38.0)
		tmp = 0.0 - (y * x);
	else
		tmp = log(2.0) + (x * (0.5 + ((x * 0.125) - y)));
	end
	tmp_2 = tmp;
end

Wolfram

code[x_, y_] := If[LessEqual[x, -38.0], N[(0.0 - N[(y * x), $MachinePrecision]), $MachinePrecision], N[(N[Log[2.0], $MachinePrecision] + N[(x * N[(0.5 + N[(N[(x * 0.125), $MachinePrecision] - y), $MachinePrecision]), $MachinePrecision]), $MachinePrecision]), $MachinePrecision]]

Derivation

1. Split input into 2 regimes

2. if x < -38

   1. Initial program 99.4%

      \[\log \left(1 + e^{x}\right) - x \cdot y \]
   2. Add Preprocessing

   3. Taylor expanded in x around inf

      \[\leadsto \color{blue}{-1 \cdot \left(x \cdot y\right)} \]
   4. Step-by-step derivation

      1. mul-1-neg N/A

         \[\leadsto \mathsf{neg}\left(x \cdot y\right) \]

      2. neg-sub0 N/A

         \[\leadsto 0 - \color{blue}{x \cdot y} \]

      3. --lowering--.f64 N/A

         \[\leadsto \mathsf{\_.f64}\left(0, \color{blue}{\left(x \cdot y\right)}\right) \]

      4. *-lowering-*.f64 99.4%

         \[\leadsto \mathsf{\_.f64}\left(0, \mathsf{*.f64}\left(x, \color{blue}{y}\right)\right) \]

   5. Simplified 99.4%

      \[\leadsto \color{blue}{0 - x \cdot y} \]

   7. Applied egg-rr 99.4%

      \[\leadsto \color{blue}{-y \cdot x} \]

   if -38 < x

   1. Initial program 98.8%

      \[\log \left(1 + e^{x}\right) - x \cdot y \]
   2. Add Preprocessing

   3. Taylor expanded in x around 0

      \[\leadsto \color{blue}{\log 2 + x \cdot \left(\left(\frac{1}{2} + \frac{1}{8} \cdot x\right) - y\right)} \]
   4. Step-by-step derivation

      1. +-lowering-+.f64 N/A

         \[\leadsto \mathsf{+.f64}\left(\log 2, \color{blue}{\left(x \cdot \left(\left(\frac{1}{2} + \frac{1}{8} \cdot x\right) - y\right)\right)}\right) \]

      2. log-lowering-log.f64 N/A

         \[\leadsto \mathsf{+.f64}\left(\mathsf{log.f64}\left(2\right), \left(\color{blue}{x} \cdot \left(\left(\frac{1}{2} + \frac{1}{8} \cdot x\right) - y\right)\right)\right) \]

      3. *-lowering-*.f64 N/A

         \[\leadsto \mathsf{+.f64}\left(\mathsf{log.f64}\left(2\right), \mathsf{*.f64}\left(x, \color{blue}{\left(\left(\frac{1}{2} + \frac{1}{8} \cdot x\right) - y\right)}\right)\right) \]

      4. associate--l+ N/A

         \[\leadsto \mathsf{+.f64}\left(\mathsf{log.f64}\left(2\right), \mathsf{*.f64}\left(x, \left(\frac{1}{2} + \color{blue}{\left(\frac{1}{8} \cdot x - y\right)}\right)\right)\right) \]

      5. +-lowering-+.f64 N/A

         \[\leadsto \mathsf{+.f64}\left(\mathsf{log.f64}\left(2\right), \mathsf{*.f64}\left(x, \mathsf{+.f64}\left(\frac{1}{2}, \color{blue}{\left(\frac{1}{8} \cdot x - y\right)}\right)\right)\right) \]

      6. --lowering--.f64 N/A

         \[\leadsto \mathsf{+.f64}\left(\mathsf{log.f64}\left(2\right), \mathsf{*.f64}\left(x, \mathsf{+.f64}\left(\frac{1}{2}, \mathsf{\_.f64}\left(\left(\frac{1}{8} \cdot x\right), \color{blue}{y}\right)\right)\right)\right) \]

      7. *-commutative N/A

         \[\leadsto \mathsf{+.f64}\left(\mathsf{log.f64}\left(2\right), \mathsf{*.f64}\left(x, \mathsf{+.f64}\left(\frac{1}{2}, \mathsf{\_.f64}\left(\left(x \cdot \frac{1}{8}\right), y\right)\right)\right)\right) \]

      8. *-lowering-*.f64 99.3%

         \[\leadsto \mathsf{+.f64}\left(\mathsf{log.f64}\left(2\right), \mathsf{*.f64}\left(x, \mathsf{+.f64}\left(\frac{1}{2}, \mathsf{\_.f64}\left(\mathsf{*.f64}\left(x, \frac{1}{8}\right), y\right)\right)\right)\right) \]

   5. Simplified 99.3%

      \[\leadsto \color{blue}{\log 2 + x \cdot \left(0.5 + \left(x \cdot 0.125 - y\right)\right)} \]
3. Recombined 2 regimes into one program.

4. Final simplification 99.4%

   \[\leadsto \begin{array}{l} \mathbf{if}\;x \leq -38:\\ \;\;\;\;0 - y \cdot x\\ \mathbf{else}:\\ \;\;\;\;\log 2 + x \cdot \left(0.5 + \left(x \cdot 0.125 - y\right)\right)\\ \end{array} \]
5. Add Preprocessing

Alternative 5: 99.2% accurate, 1.8× speedup

Math

\[\begin{array}{l} \\ \begin{array}{l} \mathbf{if}\;x \leq -1.38:\\ \;\;\;\;0 - y \cdot x\\ \mathbf{else}:\\ \;\;\;\;\log 2 + x \cdot \left(0.5 - y\right)\\ \end{array} \end{array} \]

FPCore

(FPCore (x y)
 :precision binary64
 (if (<= x -1.38) (- 0.0 (* y x)) (+ (log 2.0) (* x (- 0.5 y)))))

C

double code(double x, double y) {
	double tmp;
	if (x <= -1.38) {
		tmp = 0.0 - (y * x);
	} else {
		tmp = log(2.0) + (x * (0.5 - y));
	}
	return tmp;
}

Fortran

real(8) function code(x, y)
    real(8), intent (in) :: x
    real(8), intent (in) :: y
    real(8) :: tmp
    if (x <= (-1.38d0)) then
        tmp = 0.0d0 - (y * x)
    else
        tmp = log(2.0d0) + (x * (0.5d0 - y))
    end if
    code = tmp
end function

Java

public static double code(double x, double y) {
	double tmp;
	if (x <= -1.38) {
		tmp = 0.0 - (y * x);
	} else {
		tmp = Math.log(2.0) + (x * (0.5 - y));
	}
	return tmp;
}

Python

def code(x, y):
	tmp = 0
	if x <= -1.38:
		tmp = 0.0 - (y * x)
	else:
		tmp = math.log(2.0) + (x * (0.5 - y))
	return tmp

Julia

function code(x, y)
	tmp = 0.0
	if (x <= -1.38)
		tmp = Float64(0.0 - Float64(y * x));
	else
		tmp = Float64(log(2.0) + Float64(x * Float64(0.5 - y)));
	end
	return tmp
end

MATLAB

function tmp_2 = code(x, y)
	tmp = 0.0;
	if (x <= -1.38)
		tmp = 0.0 - (y * x);
	else
		tmp = log(2.0) + (x * (0.5 - y));
	end
	tmp_2 = tmp;
end

Wolfram

code[x_, y_] := If[LessEqual[x, -1.38], N[(0.0 - N[(y * x), $MachinePrecision]), $MachinePrecision], N[(N[Log[2.0], $MachinePrecision] + N[(x * N[(0.5 - y), $MachinePrecision]), $MachinePrecision]), $MachinePrecision]]

Derivation

1. Split input into 2 regimes

2. if x < -1.3799999999999999

   1. Initial program 99.4%

      \[\log \left(1 + e^{x}\right) - x \cdot y \]
   2. Add Preprocessing

   3. Taylor expanded in x around inf

      \[\leadsto \color{blue}{-1 \cdot \left(x \cdot y\right)} \]
   4. Step-by-step derivation

      1. mul-1-neg N/A

         \[\leadsto \mathsf{neg}\left(x \cdot y\right) \]

      2. neg-sub0 N/A

         \[\leadsto 0 - \color{blue}{x \cdot y} \]

      3. --lowering--.f64 N/A

         \[\leadsto \mathsf{\_.f64}\left(0, \color{blue}{\left(x \cdot y\right)}\right) \]

      4. *-lowering-*.f64 99.4%

         \[\leadsto \mathsf{\_.f64}\left(0, \mathsf{*.f64}\left(x, \color{blue}{y}\right)\right) \]

   5. Simplified 99.4%

      \[\leadsto \color{blue}{0 - x \cdot y} \]

   7. Applied egg-rr 99.4%

      \[\leadsto \color{blue}{-y \cdot x} \]

   if -1.3799999999999999 < x

   1. Initial program 98.8%

      \[\log \left(1 + e^{x}\right) - x \cdot y \]
   2. Add Preprocessing

   3. Taylor expanded in x around 0

      \[\leadsto \color{blue}{\log 2 + x \cdot \left(\frac{1}{2} - y\right)} \]
   4. Step-by-step derivation

      1. +-lowering-+.f64 N/A

         \[\leadsto \mathsf{+.f64}\left(\log 2, \color{blue}{\left(x \cdot \left(\frac{1}{2} - y\right)\right)}\right) \]

      2. log-lowering-log.f64 N/A

         \[\leadsto \mathsf{+.f64}\left(\mathsf{log.f64}\left(2\right), \left(\color{blue}{x} \cdot \left(\frac{1}{2} - y\right)\right)\right) \]

      3. *-lowering-*.f64 N/A

         \[\leadsto \mathsf{+.f64}\left(\mathsf{log.f64}\left(2\right), \mathsf{*.f64}\left(x, \color{blue}{\left(\frac{1}{2} - y\right)}\right)\right) \]

      4. --lowering--.f64 99.2%

         \[\leadsto \mathsf{+.f64}\left(\mathsf{log.f64}\left(2\right), \mathsf{*.f64}\left(x, \mathsf{\_.f64}\left(\frac{1}{2}, \color{blue}{y}\right)\right)\right) \]

   5. Simplified 99.2%

      \[\leadsto \color{blue}{\log 2 + x \cdot \left(0.5 - y\right)} \]
3. Recombined 2 regimes into one program.

4. Final simplification 99.3%

   \[\leadsto \begin{array}{l} \mathbf{if}\;x \leq -1.38:\\ \;\;\;\;0 - y \cdot x\\ \mathbf{else}:\\ \;\;\;\;\log 2 + x \cdot \left(0.5 - y\right)\\ \end{array} \]
5. Add Preprocessing

Alternative 6: 82.1% accurate, 1.9× speedup

Math

\[\begin{array}{l} \\ \begin{array}{l} t_0 := 0 - y \cdot x\\ \mathbf{if}\;x \leq -3.2 \cdot 10^{-76}:\\ \;\;\;\;t\_0\\ \mathbf{elif}\;x \leq 2.1 \cdot 10^{-35}:\\ \;\;\;\;\log 2\\ \mathbf{else}:\\ \;\;\;\;t\_0\\ \end{array} \end{array} \]

FPCore

(FPCore (x y)
 :precision binary64
 (let* ((t_0 (- 0.0 (* y x))))
   (if (<= x -3.2e-76) t_0 (if (<= x 2.1e-35) (log 2.0) t_0))))

C

double code(double x, double y) {
	double t_0 = 0.0 - (y * x);
	double tmp;
	if (x <= -3.2e-76) {
		tmp = t_0;
	} else if (x <= 2.1e-35) {
		tmp = log(2.0);
	} else {
		tmp = t_0;
	}
	return tmp;
}

Fortran

real(8) function code(x, y)
    real(8), intent (in) :: x
    real(8), intent (in) :: y
    real(8) :: t_0
    real(8) :: tmp
    t_0 = 0.0d0 - (y * x)
    if (x <= (-3.2d-76)) then
        tmp = t_0
    else if (x <= 2.1d-35) then
        tmp = log(2.0d0)
    else
        tmp = t_0
    end if
    code = tmp
end function

Java

public static double code(double x, double y) {
	double t_0 = 0.0 - (y * x);
	double tmp;
	if (x <= -3.2e-76) {
		tmp = t_0;
	} else if (x <= 2.1e-35) {
		tmp = Math.log(2.0);
	} else {
		tmp = t_0;
	}
	return tmp;
}

Python

def code(x, y):
	t_0 = 0.0 - (y * x)
	tmp = 0
	if x <= -3.2e-76:
		tmp = t_0
	elif x <= 2.1e-35:
		tmp = math.log(2.0)
	else:
		tmp = t_0
	return tmp

Julia

function code(x, y)
	t_0 = Float64(0.0 - Float64(y * x))
	tmp = 0.0
	if (x <= -3.2e-76)
		tmp = t_0;
	elseif (x <= 2.1e-35)
		tmp = log(2.0);
	else
		tmp = t_0;
	end
	return tmp
end

MATLAB

function tmp_2 = code(x, y)
	t_0 = 0.0 - (y * x);
	tmp = 0.0;
	if (x <= -3.2e-76)
		tmp = t_0;
	elseif (x <= 2.1e-35)
		tmp = log(2.0);
	else
		tmp = t_0;
	end
	tmp_2 = tmp;
end

Wolfram

code[x_, y_] := Block[{t$95$0 = N[(0.0 - N[(y * x), $MachinePrecision]), $MachinePrecision]}, If[LessEqual[x, -3.2e-76], t$95$0, If[LessEqual[x, 2.1e-35], N[Log[2.0], $MachinePrecision], t$95$0]]]

Derivation

1. Split input into 2 regimes

2. if x < -3.1999999999999998e-76 or 2.1e-35 < x

   1. Initial program 97.9%

      \[\log \left(1 + e^{x}\right) - x \cdot y \]
   2. Add Preprocessing

   3. Taylor expanded in x around inf

      \[\leadsto \color{blue}{-1 \cdot \left(x \cdot y\right)} \]
   4. Step-by-step derivation

      1. mul-1-neg N/A

         \[\leadsto \mathsf{neg}\left(x \cdot y\right) \]

      2. neg-sub0 N/A

         \[\leadsto 0 - \color{blue}{x \cdot y} \]

      3. --lowering--.f64 N/A

         \[\leadsto \mathsf{\_.f64}\left(0, \color{blue}{\left(x \cdot y\right)}\right) \]

      4. *-lowering-*.f64 87.9%

         \[\leadsto \mathsf{\_.f64}\left(0, \mathsf{*.f64}\left(x, \color{blue}{y}\right)\right) \]

   5. Simplified 87.9%

      \[\leadsto \color{blue}{0 - x \cdot y} \]

   7. Applied egg-rr 87.9%

      \[\leadsto \color{blue}{-y \cdot x} \]

   if -3.1999999999999998e-76 < x < 2.1e-35

   1. Initial program 100.0%

      \[\log \left(1 + e^{x}\right) - x \cdot y \]
   2. Add Preprocessing

   3. Taylor expanded in x around 0

      \[\leadsto \color{blue}{\log 2} \]
   4. Step-by-step derivation

      1. log-lowering-log.f64 83.1%

         \[\leadsto \mathsf{log.f64}\left(2\right) \]

   5. Simplified 83.1%

      \[\leadsto \color{blue}{\log 2} \]
3. Recombined 2 regimes into one program.

4. Final simplification 85.3%

   \[\leadsto \begin{array}{l} \mathbf{if}\;x \leq -3.2 \cdot 10^{-76}:\\ \;\;\;\;0 - y \cdot x\\ \mathbf{elif}\;x \leq 2.1 \cdot 10^{-35}:\\ \;\;\;\;\log 2\\ \mathbf{else}:\\ \;\;\;\;0 - y \cdot x\\ \end{array} \]
5. Add Preprocessing

Alternative 7: 98.9% accurate, 1.9× speedup

Math

\[\begin{array}{l} \\ \begin{array}{l} \mathbf{if}\;x \leq -52:\\ \;\;\;\;0 - y \cdot x\\ \mathbf{else}:\\ \;\;\;\;\mathsf{log1p}\left(1\right) - y \cdot x\\ \end{array} \end{array} \]

FPCore

(FPCore (x y)
 :precision binary64
 (if (<= x -52.0) (- 0.0 (* y x)) (- (log1p 1.0) (* y x))))

C

double code(double x, double y) {
	double tmp;
	if (x <= -52.0) {
		tmp = 0.0 - (y * x);
	} else {
		tmp = log1p(1.0) - (y * x);
	}
	return tmp;
}

Java

public static double code(double x, double y) {
	double tmp;
	if (x <= -52.0) {
		tmp = 0.0 - (y * x);
	} else {
		tmp = Math.log1p(1.0) - (y * x);
	}
	return tmp;
}

Python

def code(x, y):
	tmp = 0
	if x <= -52.0:
		tmp = 0.0 - (y * x)
	else:
		tmp = math.log1p(1.0) - (y * x)
	return tmp

Julia

function code(x, y)
	tmp = 0.0
	if (x <= -52.0)
		tmp = Float64(0.0 - Float64(y * x));
	else
		tmp = Float64(log1p(1.0) - Float64(y * x));
	end
	return tmp
end

Wolfram

code[x_, y_] := If[LessEqual[x, -52.0], N[(0.0 - N[(y * x), $MachinePrecision]), $MachinePrecision], N[(N[Log[1 + 1.0], $MachinePrecision] - N[(y * x), $MachinePrecision]), $MachinePrecision]]

Derivation

1. Split input into 2 regimes

2. if x < -52

   1. Initial program 99.4%

      \[\log \left(1 + e^{x}\right) - x \cdot y \]
   2. Add Preprocessing

   3. Taylor expanded in x around inf

      \[\leadsto \color{blue}{-1 \cdot \left(x \cdot y\right)} \]
   4. Step-by-step derivation

      1. mul-1-neg N/A

         \[\leadsto \mathsf{neg}\left(x \cdot y\right) \]

      2. neg-sub0 N/A

         \[\leadsto 0 - \color{blue}{x \cdot y} \]

      3. --lowering--.f64 N/A

         \[\leadsto \mathsf{\_.f64}\left(0, \color{blue}{\left(x \cdot y\right)}\right) \]

      4. *-lowering-*.f64 99.4%

         \[\leadsto \mathsf{\_.f64}\left(0, \mathsf{*.f64}\left(x, \color{blue}{y}\right)\right) \]

   5. Simplified 99.4%

      \[\leadsto \color{blue}{0 - x \cdot y} \]

   7. Applied egg-rr 99.4%

      \[\leadsto \color{blue}{-y \cdot x} \]

   if -52 < x

   1. Initial program 98.8%

      \[\log \left(1 + e^{x}\right) - x \cdot y \]
   2. Step-by-step derivation

      1. --lowering--.f64 N/A

         \[\leadsto \mathsf{\_.f64}\left(\log \left(1 + e^{x}\right), \color{blue}{\left(x \cdot y\right)}\right) \]

      2. log1p-define N/A

         \[\leadsto \mathsf{\_.f64}\left(\left(\mathsf{log1p}\left(e^{x}\right)\right), \left(\color{blue}{x} \cdot y\right)\right) \]

      3. log1p-lowering-log1p.f64 N/A

         \[\leadsto \mathsf{\_.f64}\left(\mathsf{log1p.f64}\left(\left(e^{x}\right)\right), \left(\color{blue}{x} \cdot y\right)\right) \]

      4. exp-lowering-exp.f64 N/A

         \[\leadsto \mathsf{\_.f64}\left(\mathsf{log1p.f64}\left(\mathsf{exp.f64}\left(x\right)\right), \left(x \cdot y\right)\right) \]

      5. *-lowering-*.f64 98.8%

         \[\leadsto \mathsf{\_.f64}\left(\mathsf{log1p.f64}\left(\mathsf{exp.f64}\left(x\right)\right), \mathsf{*.f64}\left(x, \color{blue}{y}\right)\right) \]

   3. Simplified 98.8%

      \[\leadsto \color{blue}{\mathsf{log1p}\left(e^{x}\right) - x \cdot y} \]
   4. Add Preprocessing

   5. Taylor expanded in x around 0

      \[\leadsto \mathsf{\_.f64}\left(\mathsf{log1p.f64}\left(\color{blue}{1}\right), \mathsf{*.f64}\left(x, y\right)\right) \]
   6. Step-by-step derivation

   7. Simplified 98.2%

      \[\leadsto \mathsf{log1p}\left(\color{blue}{1}\right) - x \cdot y \]
3. Recombined 2 regimes into one program.

4. Final simplification 98.6%

   \[\leadsto \begin{array}{l} \mathbf{if}\;x \leq -52:\\ \;\;\;\;0 - y \cdot x\\ \mathbf{else}:\\ \;\;\;\;\mathsf{log1p}\left(1\right) - y \cdot x\\ \end{array} \]
5. Add Preprocessing

Alternative 8: 50.8% accurate, 41.4× speedup

Math

\[\begin{array}{l} \\ 0 - y \cdot x \end{array} \]

FPCore

(FPCore (x y) :precision binary64 (- 0.0 (* y x)))

C

double code(double x, double y) {
	return 0.0 - (y * x);
}

Fortran

real(8) function code(x, y)
    real(8), intent (in) :: x
    real(8), intent (in) :: y
    code = 0.0d0 - (y * x)
end function

Java

public static double code(double x, double y) {
	return 0.0 - (y * x);
}

Python

def code(x, y):
	return 0.0 - (y * x)

Julia

function code(x, y)
	return Float64(0.0 - Float64(y * x))
end

MATLAB

function tmp = code(x, y)
	tmp = 0.0 - (y * x);
end

Wolfram

code[x_, y_] := N[(0.0 - N[(y * x), $MachinePrecision]), $MachinePrecision]

Derivation

1. Initial program 99.0%

   \[\log \left(1 + e^{x}\right) - x \cdot y \]
2. Add Preprocessing

3. Taylor expanded in x around inf

   \[\leadsto \color{blue}{-1 \cdot \left(x \cdot y\right)} \]
4. Step-by-step derivation

   1. mul-1-neg N/A

      \[\leadsto \mathsf{neg}\left(x \cdot y\right) \]

   2. neg-sub0 N/A

      \[\leadsto 0 - \color{blue}{x \cdot y} \]

   3. --lowering--.f64 N/A

      \[\leadsto \mathsf{\_.f64}\left(0, \color{blue}{\left(x \cdot y\right)}\right) \]

   4. *-lowering-*.f64 50.3%

      \[\leadsto \mathsf{\_.f64}\left(0, \mathsf{*.f64}\left(x, \color{blue}{y}\right)\right) \]

5. Simplified 50.3%

   \[\leadsto \color{blue}{0 - x \cdot y} \]

7. Applied egg-rr 50.3%

   \[\leadsto \color{blue}{-y \cdot x} \]

8. Final simplification 50.3%

   \[\leadsto 0 - y \cdot x \]
9. Add Preprocessing

Alternative 9: 2.3% accurate, 69.0× speedup

Math

\[\begin{array}{l} \\ y \cdot x \end{array} \]

FPCore

(FPCore (x y) :precision binary64 (* y x))

C

double code(double x, double y) {
	return y * x;
}

Fortran

real(8) function code(x, y)
    real(8), intent (in) :: x
    real(8), intent (in) :: y
    code = y * x
end function

Java

public static double code(double x, double y) {
	return y * x;
}

Python

def code(x, y):
	return y * x

Julia

function code(x, y)
	return Float64(y * x)
end

MATLAB

function tmp = code(x, y)
	tmp = y * x;
end

Wolfram

code[x_, y_] := N[(y * x), $MachinePrecision]

Derivation

1. Initial program 99.0%

   \[\log \left(1 + e^{x}\right) - x \cdot y \]
2. Add Preprocessing

3. Taylor expanded in x around inf

   \[\leadsto \color{blue}{-1 \cdot \left(x \cdot y\right)} \]
4. Step-by-step derivation

   1. mul-1-neg N/A

      \[\leadsto \mathsf{neg}\left(x \cdot y\right) \]

   2. neg-sub0 N/A

      \[\leadsto 0 - \color{blue}{x \cdot y} \]

   3. --lowering--.f64 N/A

      \[\leadsto \mathsf{\_.f64}\left(0, \color{blue}{\left(x \cdot y\right)}\right) \]

   4. *-lowering-*.f64 50.3%

      \[\leadsto \mathsf{\_.f64}\left(0, \mathsf{*.f64}\left(x, \color{blue}{y}\right)\right) \]

5. Simplified 50.3%

   \[\leadsto \color{blue}{0 - x \cdot y} \]
6. Step-by-step derivation

   1. flip3-- N/A

      \[\leadsto \frac{{0}^{3} - {\left(x \cdot y\right)}^{3}}{\color{blue}{0 \cdot 0 + \left(\left(x \cdot y\right) \cdot \left(x \cdot y\right) + 0 \cdot \left(x \cdot y\right)\right)}} \]

7. Applied egg-rr 2.2%

   \[\leadsto \color{blue}{y \cdot x} \]
8. Add Preprocessing

Developer Target 1: 99.9% accurate, 1.0× speedup

Math

\[\begin{array}{l} \\ \begin{array}{l} \mathbf{if}\;x \leq 0:\\ \;\;\;\;\log \left(1 + e^{x}\right) - x \cdot y\\ \mathbf{else}:\\ \;\;\;\;\log \left(1 + e^{-x}\right) - \left(-x\right) \cdot \left(1 - y\right)\\ \end{array} \end{array} \]

FPCore

(FPCore (x y)
 :precision binary64
 (if (<= x 0.0)
   (- (log (+ 1.0 (exp x))) (* x y))
   (- (log (+ 1.0 (exp (- x)))) (* (- x) (- 1.0 y)))))

C

double code(double x, double y) {
	double tmp;
	if (x <= 0.0) {
		tmp = log((1.0 + exp(x))) - (x * y);
	} else {
		tmp = log((1.0 + exp(-x))) - (-x * (1.0 - y));
	}
	return tmp;
}

Fortran

real(8) function code(x, y)
    real(8), intent (in) :: x
    real(8), intent (in) :: y
    real(8) :: tmp
    if (x <= 0.0d0) then
        tmp = log((1.0d0 + exp(x))) - (x * y)
    else
        tmp = log((1.0d0 + exp(-x))) - (-x * (1.0d0 - y))
    end if
    code = tmp
end function

Java

public static double code(double x, double y) {
	double tmp;
	if (x <= 0.0) {
		tmp = Math.log((1.0 + Math.exp(x))) - (x * y);
	} else {
		tmp = Math.log((1.0 + Math.exp(-x))) - (-x * (1.0 - y));
	}
	return tmp;
}

Python

def code(x, y):
	tmp = 0
	if x <= 0.0:
		tmp = math.log((1.0 + math.exp(x))) - (x * y)
	else:
		tmp = math.log((1.0 + math.exp(-x))) - (-x * (1.0 - y))
	return tmp

Julia

function code(x, y)
	tmp = 0.0
	if (x <= 0.0)
		tmp = Float64(log(Float64(1.0 + exp(x))) - Float64(x * y));
	else
		tmp = Float64(log(Float64(1.0 + exp(Float64(-x)))) - Float64(Float64(-x) * Float64(1.0 - y)));
	end
	return tmp
end

MATLAB

function tmp_2 = code(x, y)
	tmp = 0.0;
	if (x <= 0.0)
		tmp = log((1.0 + exp(x))) - (x * y);
	else
		tmp = log((1.0 + exp(-x))) - (-x * (1.0 - y));
	end
	tmp_2 = tmp;
end

Wolfram

code[x_, y_] := If[LessEqual[x, 0.0], N[(N[Log[N[(1.0 + N[Exp[x], $MachinePrecision]), $MachinePrecision]], $MachinePrecision] - N[(x * y), $MachinePrecision]), $MachinePrecision], N[(N[Log[N[(1.0 + N[Exp[(-x)], $MachinePrecision]), $MachinePrecision]], $MachinePrecision] - N[((-x) * N[(1.0 - y), $MachinePrecision]), $MachinePrecision]), $MachinePrecision]]

Reproduce

herbie shell --seed 2024150 
(FPCore (x y)
  :name "Logistic regression 2"
  :precision binary64

  :alt
  (! :herbie-platform default (if (<= x 0) (- (log (+ 1 (exp x))) (* x y)) (- (log (+ 1 (exp (- x)))) (* (- x) (- 1 y)))))

  (- (log (+ 1.0 (exp x))) (* x y)))