Logistic function from Lakshay Garg

Average Error: 29.1 → 0.1

Time: 4.2s

Precision: binary64

Math

\[\frac{2}{1 + e^{-2 \cdot x}} - 1\]

↓

\[\begin{array}{l} \mathbf{if}\;-2 \cdot x \leq -1.5465792801287923 \lor \neg \left(-2 \cdot x \leq 1.065631806124204 \cdot 10^{-08}\right):\\ \;\;\;\;\frac{\frac{4}{\left(1 + e^{-2 \cdot x}\right) \cdot \left(1 + e^{-2 \cdot x}\right)} + -1}{1 + \frac{2}{1 + e^{-2 \cdot x}}}\\ \mathbf{else}:\\ \;\;\;\;\left(x + 0.13333333333333333 \cdot {x}^{5}\right) - 0.3333333333333333 \cdot {x}^{3}\\ \end{array}\]

FPCore

(FPCore (x y) :precision binary64 (- (/ 2.0 (+ 1.0 (exp (* -2.0 x)))) 1.0))

↓

(FPCore (x y)
 :precision binary64
 (if (or (<= (* -2.0 x) -1.5465792801287923)
         (not (<= (* -2.0 x) 1.065631806124204e-08)))
   (/
    (+ (/ 4.0 (* (+ 1.0 (exp (* -2.0 x))) (+ 1.0 (exp (* -2.0 x))))) -1.0)
    (+ 1.0 (/ 2.0 (+ 1.0 (exp (* -2.0 x))))))
   (-
    (+ x (* 0.13333333333333333 (pow x 5.0)))
    (* 0.3333333333333333 (pow x 3.0)))))

C

double code(double x, double y) {
	return (2.0 / (1.0 + exp(-2.0 * x))) - 1.0;
}

↓

double code(double x, double y) {
	double tmp;
	if (((-2.0 * x) <= -1.5465792801287923) || !((-2.0 * x) <= 1.065631806124204e-08)) {
		tmp = ((4.0 / ((1.0 + exp(-2.0 * x)) * (1.0 + exp(-2.0 * x)))) + -1.0) / (1.0 + (2.0 / (1.0 + exp(-2.0 * x))));
	} else {
		tmp = (x + (0.13333333333333333 * pow(x, 5.0))) - (0.3333333333333333 * pow(x, 3.0));
	}
	return tmp;
}

Error

Bits error versus x

Bits error versus y

Derivation

1. Split input into 2 regimes

2. if (*.f64 -2 x) < -1.54657928012879231 or 1.06563180612420405e-8 < (*.f64 -2 x)

   1. Initial program 0.1

      \[\frac{2}{1 + e^{-2 \cdot x}} - 1\]
   2. Using strategy rm

   3. Applied flip--_binary64_742 0.1

      \[\leadsto \color{blue}{\frac{\frac{2}{1 + e^{-2 \cdot x}} \cdot \frac{2}{1 + e^{-2 \cdot x}} - 1 \cdot 1}{\frac{2}{1 + e^{-2 \cdot x}} + 1}}\]

   4. Simplified 0.1

      \[\leadsto \frac{\color{blue}{\frac{4}{\left(1 + e^{-2 \cdot x}\right) \cdot \left(1 + e^{-2 \cdot x}\right)} + -1}}{\frac{2}{1 + e^{-2 \cdot x}} + 1}\]

   5. Simplified 0.1

      \[\leadsto \frac{\frac{4}{\left(1 + e^{-2 \cdot x}\right) \cdot \left(1 + e^{-2 \cdot x}\right)} + -1}{\color{blue}{1 + \frac{2}{1 + e^{-2 \cdot x}}}}\]

   if -1.54657928012879231 < (*.f64 -2 x) < 1.06563180612420405e-8

   1. Initial program 59.2

      \[\frac{2}{1 + e^{-2 \cdot x}} - 1\]

   2. Taylor expanded around 0 0.1

      \[\leadsto \color{blue}{\left(x + 0.13333333333333333 \cdot {x}^{5}\right) - 0.3333333333333333 \cdot {x}^{3}}\]
3. Recombined 2 regimes into one program.

4. Final simplification 0.1

   \[\leadsto \begin{array}{l} \mathbf{if}\;-2 \cdot x \leq -1.5465792801287923 \lor \neg \left(-2 \cdot x \leq 1.065631806124204 \cdot 10^{-08}\right):\\ \;\;\;\;\frac{\frac{4}{\left(1 + e^{-2 \cdot x}\right) \cdot \left(1 + e^{-2 \cdot x}\right)} + -1}{1 + \frac{2}{1 + e^{-2 \cdot x}}}\\ \mathbf{else}:\\ \;\;\;\;\left(x + 0.13333333333333333 \cdot {x}^{5}\right) - 0.3333333333333333 \cdot {x}^{3}\\ \end{array}\]

Reproduce

herbie shell --seed 2020292 
(FPCore (x y)
  :name "Logistic function from Lakshay Garg"
  :precision binary64
  (- (/ 2.0 (+ 1.0 (exp (* -2.0 x)))) 1.0))