Raymond Hettinger | 6fee0f8 | 2019-08-26 11:25:58 -0700 | [diff] [blame] | 1 | /* statistics accelerator C extension: _statistics module. */ |
Dong-hee Na | 0a18ee4 | 2019-08-24 07:20:30 +0900 | [diff] [blame] | 2 | |
| 3 | #include "Python.h" |
| 4 | #include "structmember.h" |
| 5 | #include "clinic/_statisticsmodule.c.h" |
| 6 | |
| 7 | /*[clinic input] |
| 8 | module _statistics |
| 9 | |
| 10 | [clinic start generated code]*/ |
| 11 | /*[clinic end generated code: output=da39a3ee5e6b4b0d input=864a6f59b76123b2]*/ |
| 12 | |
Raymond Hettinger | 6fee0f8 | 2019-08-26 11:25:58 -0700 | [diff] [blame] | 13 | /* |
| 14 | * There is no closed-form solution to the inverse CDF for the normal |
| 15 | * distribution, so we use a rational approximation instead: |
| 16 | * Wichura, M.J. (1988). "Algorithm AS241: The Percentage Points of the |
| 17 | * Normal Distribution". Applied Statistics. Blackwell Publishing. 37 |
| 18 | * (3): 477–484. doi:10.2307/2347330. JSTOR 2347330. |
| 19 | */ |
Dong-hee Na | 0a18ee4 | 2019-08-24 07:20:30 +0900 | [diff] [blame] | 20 | |
| 21 | /*[clinic input] |
| 22 | _statistics._normal_dist_inv_cdf -> double |
| 23 | p: double |
| 24 | mu: double |
| 25 | sigma: double |
| 26 | / |
| 27 | [clinic start generated code]*/ |
| 28 | |
| 29 | static double |
| 30 | _statistics__normal_dist_inv_cdf_impl(PyObject *module, double p, double mu, |
| 31 | double sigma) |
| 32 | /*[clinic end generated code: output=02fd19ddaab36602 input=24715a74be15296a]*/ |
| 33 | { |
| 34 | double q, num, den, r, x; |
| 35 | q = p - 0.5; |
| 36 | // Algorithm AS 241: The Percentage Points of the Normal Distribution |
| 37 | if(fabs(q) <= 0.425) { |
| 38 | r = 0.180625 - q * q; |
Raymond Hettinger | 6fee0f8 | 2019-08-26 11:25:58 -0700 | [diff] [blame] | 39 | // Hash sum-55.8831928806149014439 |
Dong-hee Na | 0a18ee4 | 2019-08-24 07:20:30 +0900 | [diff] [blame] | 40 | num = (((((((2.5090809287301226727e+3 * r + |
| 41 | 3.3430575583588128105e+4) * r + |
| 42 | 6.7265770927008700853e+4) * r + |
| 43 | 4.5921953931549871457e+4) * r + |
| 44 | 1.3731693765509461125e+4) * r + |
| 45 | 1.9715909503065514427e+3) * r + |
| 46 | 1.3314166789178437745e+2) * r + |
| 47 | 3.3871328727963666080e+0) * q; |
| 48 | den = (((((((5.2264952788528545610e+3 * r + |
| 49 | 2.8729085735721942674e+4) * r + |
| 50 | 3.9307895800092710610e+4) * r + |
| 51 | 2.1213794301586595867e+4) * r + |
| 52 | 5.3941960214247511077e+3) * r + |
| 53 | 6.8718700749205790830e+2) * r + |
| 54 | 4.2313330701600911252e+1) * r + |
| 55 | 1.0); |
| 56 | x = num / den; |
| 57 | return mu + (x * sigma); |
| 58 | } |
Raymond Hettinger | 6fee0f8 | 2019-08-26 11:25:58 -0700 | [diff] [blame] | 59 | r = (q <= 0.0) ? p : (1.0 - p); |
Dong-hee Na | 0a18ee4 | 2019-08-24 07:20:30 +0900 | [diff] [blame] | 60 | r = sqrt(-log(r)); |
| 61 | if (r <= 5.0) { |
| 62 | r = r - 1.6; |
Raymond Hettinger | 6fee0f8 | 2019-08-26 11:25:58 -0700 | [diff] [blame] | 63 | // Hash sum-49.33206503301610289036 |
Dong-hee Na | 0a18ee4 | 2019-08-24 07:20:30 +0900 | [diff] [blame] | 64 | num = (((((((7.74545014278341407640e-4 * r + |
| 65 | 2.27238449892691845833e-2) * r + |
| 66 | 2.41780725177450611770e-1) * r + |
| 67 | 1.27045825245236838258e+0) * r + |
| 68 | 3.64784832476320460504e+0) * r + |
| 69 | 5.76949722146069140550e+0) * r + |
| 70 | 4.63033784615654529590e+0) * r + |
| 71 | 1.42343711074968357734e+0); |
| 72 | den = (((((((1.05075007164441684324e-9 * r + |
| 73 | 5.47593808499534494600e-4) * r + |
| 74 | 1.51986665636164571966e-2) * r + |
| 75 | 1.48103976427480074590e-1) * r + |
| 76 | 6.89767334985100004550e-1) * r + |
| 77 | 1.67638483018380384940e+0) * r + |
| 78 | 2.05319162663775882187e+0) * r + |
| 79 | 1.0); |
| 80 | } else { |
| 81 | r -= 5.0; |
Raymond Hettinger | 6fee0f8 | 2019-08-26 11:25:58 -0700 | [diff] [blame] | 82 | // Hash sum-47.52583317549289671629 |
Dong-hee Na | 0a18ee4 | 2019-08-24 07:20:30 +0900 | [diff] [blame] | 83 | num = (((((((2.01033439929228813265e-7 * r + |
| 84 | 2.71155556874348757815e-5) * r + |
| 85 | 1.24266094738807843860e-3) * r + |
| 86 | 2.65321895265761230930e-2) * r + |
| 87 | 2.96560571828504891230e-1) * r + |
| 88 | 1.78482653991729133580e+0) * r + |
| 89 | 5.46378491116411436990e+0) * r + |
| 90 | 6.65790464350110377720e+0); |
| 91 | den = (((((((2.04426310338993978564e-15 * r + |
| 92 | 1.42151175831644588870e-7) * r + |
| 93 | 1.84631831751005468180e-5) * r + |
| 94 | 7.86869131145613259100e-4) * r + |
| 95 | 1.48753612908506148525e-2) * r + |
| 96 | 1.36929880922735805310e-1) * r + |
| 97 | 5.99832206555887937690e-1) * r + |
| 98 | 1.0); |
| 99 | } |
| 100 | x = num / den; |
Raymond Hettinger | 6fee0f8 | 2019-08-26 11:25:58 -0700 | [diff] [blame] | 101 | if (q < 0.0) { |
| 102 | x = -x; |
| 103 | } |
Dong-hee Na | 0a18ee4 | 2019-08-24 07:20:30 +0900 | [diff] [blame] | 104 | return mu + (x * sigma); |
| 105 | } |
| 106 | |
Raymond Hettinger | 6fee0f8 | 2019-08-26 11:25:58 -0700 | [diff] [blame] | 107 | |
| 108 | static PyMethodDef statistics_methods[] = { |
| 109 | _STATISTICS__NORMAL_DIST_INV_CDF_METHODDEF |
| 110 | {NULL, NULL, 0, NULL} |
| 111 | }; |
| 112 | |
Dong-hee Na | 0a18ee4 | 2019-08-24 07:20:30 +0900 | [diff] [blame] | 113 | static struct PyModuleDef statisticsmodule = { |
| 114 | PyModuleDef_HEAD_INIT, |
| 115 | "_statistics", |
| 116 | _statistics__normal_dist_inv_cdf__doc__, |
| 117 | -1, |
Raymond Hettinger | 6fee0f8 | 2019-08-26 11:25:58 -0700 | [diff] [blame] | 118 | statistics_methods, |
Dong-hee Na | 0a18ee4 | 2019-08-24 07:20:30 +0900 | [diff] [blame] | 119 | NULL, |
| 120 | NULL, |
| 121 | NULL, |
| 122 | NULL |
| 123 | }; |
| 124 | |
Dong-hee Na | 0a18ee4 | 2019-08-24 07:20:30 +0900 | [diff] [blame] | 125 | PyMODINIT_FUNC |
| 126 | PyInit__statistics(void) |
| 127 | { |
| 128 | PyObject *m = PyModule_Create(&statisticsmodule); |
| 129 | if (!m) return NULL; |
| 130 | return m; |
| 131 | } |