11\01\05\17\01S\15\03\0B\05E\07S\09U\0B[\0DA\17\011\07\03\0B\05?\07]\09?\0BC\0D_\17\01'\07\17\01)\0B\03\13#a%A'c)?+e-?/?1?3g\05\1D\05\1F\05!\05#\05%\05'\05)\05+\05-\17\013\0B\03\039C\05/\17\015\1B\17\017\0B\03\01\1D1\1D3\03\05GQ\0D\05IKMO\1D5\1D7\1D9\1D;\0D\01#\09\03\03W\0D\03YA\1D=\1D?#\0B\1DA\0B\05\1DC\05\03\0D\03ik\1DE\13\0D\01\01\02\04)\03\00\FF\FF\FF\FF\FF\FF\FF\FF\05\1B3\05\11\05\03\07\03\03\11\03\03\03\03\1D\04}\05\01\11\15\0F\07\04m\03\01\09\03\11\19\17\05\03\07\0F\05\03\03\07\03\00\07\0...

//
//
// exp(x)
// Returns the exponential of x.
//
// Method
//   1. Argument reduction:
//      Reduce x to an r so that |r| <= 0.5*ln2 ~ 0.34658.
//      Given x, find r and integer k such that
//
//               x = k*ln2 + r,  |r| <= 0.5*ln2.
//
//      Here r will be represented as r = hi-lo for better
//      accuracy.
//
//   2. Approximation of exp(r) by a special rational function on

	z := new(Zipf)
	if s <= 1.0 || v < 1 {
		return nil
	}
	z.r = r
	z.imax = float64(imax)
	z.v = v
	z.q = s
	z.oneminusQ = 1.0 - z.q
	z.oneminusQinv = 1.0 / z.oneminusQ
	z.hxm = z.h(z.imax + 0.5)
	z.hx0minusHxm = z.h(0.5) - math.Exp(math.Log(z.v)*(-z.q)) - z.hxm
	z.s = 1 - z.hinv(z.h(1.5)-math.Exp(-z.q*math.Log(z.v+1.0)))
	return z
}

// Uint64 returns a value drawn from the [Zipf] distribution described

  %0 = "tf.RecvTPUEmbeddingActivations"() {config = "\0A%\0A\0Dwatches_table\10\F5\03\18\80\02 \01*\0C\1A\00j\05\0D\00\00\80?\88\01\01\10\02\18\80\04 \01(\02"} : () -> tensor<512x256xf32>
  "tf.SendTPUEmbeddingGradients"(%0) {config = "\0A%\0A\0Dwatches_table\10\F5\03\18\80\02 \01*\0C\1A\00j\05\0D\00\00\80?\88\01\01\10\02\18\80\04 \01(\02", operandSegmentSizes = array<i32: 1, 0>} : (tensor<512x256xf32>) -> ()
  func.return
}

	// x is now in primary range
	hfx := 0.5 * x
	hxs := x * hfx
	r1 := 1 + hxs*(Q1+hxs*(Q2+hxs*(Q3+hxs*(Q4+hxs*Q5))))
	t := 3 - r1*hfx
	e := hxs * ((r1 - t) / (6.0 - x*t))
	if k == 0 {
		return x - (x*e - hxs) // c is 0
	}
	e = (x*(e-c) - c)
	e -= hxs
	switch {
	case k == -1:
		return 0.5*(x-e) - 0.5
	case k == 1:
		if x < -0.25 {
			return -2 * (e - (x + 0.5))
		}
		return 1 + 2*(x-e)

        [Insufficient.CLASS_METRIC_MISSED_COUNT]                    | 'class metric with missed count'                                         | 'classes missed count is 0.0, but expected minimum is 0.5'
        [Insufficient.LINE_METRIC_COVERED_RATIO,

	z := new(Zipf)
	if s <= 1.0 || v < 1 {
		return nil
	}
	z.r = r
	z.imax = float64(imax)
	z.v = v
	z.q = s
	z.oneminusQ = 1.0 - z.q
	z.oneminusQinv = 1.0 / z.oneminusQ
	z.hxm = z.h(z.imax + 0.5)
	z.hx0minusHxm = z.h(0.5) - math.Exp(math.Log(z.v)*(-z.q)) - z.hxm
	z.s = 1 - z.hinv(z.h(1.5)-math.Exp(-z.q*math.Log(z.v+1.0)))
	return z
}

// Uint64 returns a value drawn from the Zipf distribution described
// by the Zipf object.

		{name: "no fractional seconds", in: Date(1998, time.May, 5, 5, 5, 5, 0, time.UTC), out: []byte("\x541998-05-05T05:05:05Z")},                // '1998-05-05T05:05:05Z'
		{name: "fractional seconds truncated", in: Date(1998, time.May, 5, 5, 5, 5, 123456789, time.UTC), out: []byte("\x541998-05-05T05:05:05Z")}, // '1998-05-05T05:05:05Z'

//      is very close to 0.5.
//
//   3. For x>=8, we have
//      lgamma(x)~(x-0.5)log(x)-x+0.5*log(2pi)+1/(12x)-1/(360x**3)+....
//      (better formula:
//         lgamma(x)~(x-0.5)*(log(x)-1)-.5*(log(2pi)-1) + ...)
//      Let z = 1/x, then we approximation
//              f(z) = lgamma(x) - (x-0.5)(log(x)-1)
//      by
//                                  3       5             11

    assertEquals(-1, binarySearch(0, 5) { index -> 0.compareTo(table[index]) })
    assertEquals(-2, binarySearch(0, 5) { index -> 2.compareTo(table[index]) })
    assertEquals(-3, binarySearch(0, 5) { index -> 4.compareTo(table[index]) })
    assertEquals(-4, binarySearch(0, 5) { index -> 6.compareTo(table[index]) })
    assertEquals(-5, binarySearch(0, 5) { index -> 8.compareTo(table[index]) })