SingularValueDecomposition.php 19 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528
  1. <?php
  2. /**
  3. * @package JAMA
  4. *
  5. * For an m-by-n matrix A with m >= n, the singular value decomposition is
  6. * an m-by-n orthogonal matrix U, an n-by-n diagonal matrix S, and
  7. * an n-by-n orthogonal matrix V so that A = U*S*V'.
  8. *
  9. * The singular values, sigma[$k] = S[$k][$k], are ordered so that
  10. * sigma[0] >= sigma[1] >= ... >= sigma[n-1].
  11. *
  12. * The singular value decompostion always exists, so the constructor will
  13. * never fail. The matrix condition number and the effective numerical
  14. * rank can be computed from this decomposition.
  15. *
  16. * @author Paul Meagher
  17. * @license PHP v3.0
  18. * @version 1.1
  19. */
  20. class SingularValueDecomposition
  21. {
  22. /**
  23. * Internal storage of U.
  24. * @var array
  25. */
  26. private $U = array();
  27. /**
  28. * Internal storage of V.
  29. * @var array
  30. */
  31. private $V = array();
  32. /**
  33. * Internal storage of singular values.
  34. * @var array
  35. */
  36. private $s = array();
  37. /**
  38. * Row dimension.
  39. * @var int
  40. */
  41. private $m;
  42. /**
  43. * Column dimension.
  44. * @var int
  45. */
  46. private $n;
  47. /**
  48. * Construct the singular value decomposition
  49. *
  50. * Derived from LINPACK code.
  51. *
  52. * @param $A Rectangular matrix
  53. * @return Structure to access U, S and V.
  54. */
  55. public function __construct($Arg)
  56. {
  57. // Initialize.
  58. $A = $Arg->getArrayCopy();
  59. $this->m = $Arg->getRowDimension();
  60. $this->n = $Arg->getColumnDimension();
  61. $nu = min($this->m, $this->n);
  62. $e = array();
  63. $work = array();
  64. $wantu = true;
  65. $wantv = true;
  66. $nct = min($this->m - 1, $this->n);
  67. $nrt = max(0, min($this->n - 2, $this->m));
  68. // Reduce A to bidiagonal form, storing the diagonal elements
  69. // in s and the super-diagonal elements in e.
  70. for ($k = 0; $k < max($nct, $nrt); ++$k) {
  71. if ($k < $nct) {
  72. // Compute the transformation for the k-th column and
  73. // place the k-th diagonal in s[$k].
  74. // Compute 2-norm of k-th column without under/overflow.
  75. $this->s[$k] = 0;
  76. for ($i = $k; $i < $this->m; ++$i) {
  77. $this->s[$k] = hypo($this->s[$k], $A[$i][$k]);
  78. }
  79. if ($this->s[$k] != 0.0) {
  80. if ($A[$k][$k] < 0.0) {
  81. $this->s[$k] = -$this->s[$k];
  82. }
  83. for ($i = $k; $i < $this->m; ++$i) {
  84. $A[$i][$k] /= $this->s[$k];
  85. }
  86. $A[$k][$k] += 1.0;
  87. }
  88. $this->s[$k] = -$this->s[$k];
  89. }
  90. for ($j = $k + 1; $j < $this->n; ++$j) {
  91. if (($k < $nct) & ($this->s[$k] != 0.0)) {
  92. // Apply the transformation.
  93. $t = 0;
  94. for ($i = $k; $i < $this->m; ++$i) {
  95. $t += $A[$i][$k] * $A[$i][$j];
  96. }
  97. $t = -$t / $A[$k][$k];
  98. for ($i = $k; $i < $this->m; ++$i) {
  99. $A[$i][$j] += $t * $A[$i][$k];
  100. }
  101. // Place the k-th row of A into e for the
  102. // subsequent calculation of the row transformation.
  103. $e[$j] = $A[$k][$j];
  104. }
  105. }
  106. if ($wantu and ($k < $nct)) {
  107. // Place the transformation in U for subsequent back
  108. // multiplication.
  109. for ($i = $k; $i < $this->m; ++$i) {
  110. $this->U[$i][$k] = $A[$i][$k];
  111. }
  112. }
  113. if ($k < $nrt) {
  114. // Compute the k-th row transformation and place the
  115. // k-th super-diagonal in e[$k].
  116. // Compute 2-norm without under/overflow.
  117. $e[$k] = 0;
  118. for ($i = $k + 1; $i < $this->n; ++$i) {
  119. $e[$k] = hypo($e[$k], $e[$i]);
  120. }
  121. if ($e[$k] != 0.0) {
  122. if ($e[$k+1] < 0.0) {
  123. $e[$k] = -$e[$k];
  124. }
  125. for ($i = $k + 1; $i < $this->n; ++$i) {
  126. $e[$i] /= $e[$k];
  127. }
  128. $e[$k+1] += 1.0;
  129. }
  130. $e[$k] = -$e[$k];
  131. if (($k+1 < $this->m) and ($e[$k] != 0.0)) {
  132. // Apply the transformation.
  133. for ($i = $k+1; $i < $this->m; ++$i) {
  134. $work[$i] = 0.0;
  135. }
  136. for ($j = $k+1; $j < $this->n; ++$j) {
  137. for ($i = $k+1; $i < $this->m; ++$i) {
  138. $work[$i] += $e[$j] * $A[$i][$j];
  139. }
  140. }
  141. for ($j = $k + 1; $j < $this->n; ++$j) {
  142. $t = -$e[$j] / $e[$k+1];
  143. for ($i = $k + 1; $i < $this->m; ++$i) {
  144. $A[$i][$j] += $t * $work[$i];
  145. }
  146. }
  147. }
  148. if ($wantv) {
  149. // Place the transformation in V for subsequent
  150. // back multiplication.
  151. for ($i = $k + 1; $i < $this->n; ++$i) {
  152. $this->V[$i][$k] = $e[$i];
  153. }
  154. }
  155. }
  156. }
  157. // Set up the final bidiagonal matrix or order p.
  158. $p = min($this->n, $this->m + 1);
  159. if ($nct < $this->n) {
  160. $this->s[$nct] = $A[$nct][$nct];
  161. }
  162. if ($this->m < $p) {
  163. $this->s[$p-1] = 0.0;
  164. }
  165. if ($nrt + 1 < $p) {
  166. $e[$nrt] = $A[$nrt][$p-1];
  167. }
  168. $e[$p-1] = 0.0;
  169. // If required, generate U.
  170. if ($wantu) {
  171. for ($j = $nct; $j < $nu; ++$j) {
  172. for ($i = 0; $i < $this->m; ++$i) {
  173. $this->U[$i][$j] = 0.0;
  174. }
  175. $this->U[$j][$j] = 1.0;
  176. }
  177. for ($k = $nct - 1; $k >= 0; --$k) {
  178. if ($this->s[$k] != 0.0) {
  179. for ($j = $k + 1; $j < $nu; ++$j) {
  180. $t = 0;
  181. for ($i = $k; $i < $this->m; ++$i) {
  182. $t += $this->U[$i][$k] * $this->U[$i][$j];
  183. }
  184. $t = -$t / $this->U[$k][$k];
  185. for ($i = $k; $i < $this->m; ++$i) {
  186. $this->U[$i][$j] += $t * $this->U[$i][$k];
  187. }
  188. }
  189. for ($i = $k; $i < $this->m; ++$i) {
  190. $this->U[$i][$k] = -$this->U[$i][$k];
  191. }
  192. $this->U[$k][$k] = 1.0 + $this->U[$k][$k];
  193. for ($i = 0; $i < $k - 1; ++$i) {
  194. $this->U[$i][$k] = 0.0;
  195. }
  196. } else {
  197. for ($i = 0; $i < $this->m; ++$i) {
  198. $this->U[$i][$k] = 0.0;
  199. }
  200. $this->U[$k][$k] = 1.0;
  201. }
  202. }
  203. }
  204. // If required, generate V.
  205. if ($wantv) {
  206. for ($k = $this->n - 1; $k >= 0; --$k) {
  207. if (($k < $nrt) and ($e[$k] != 0.0)) {
  208. for ($j = $k + 1; $j < $nu; ++$j) {
  209. $t = 0;
  210. for ($i = $k + 1; $i < $this->n; ++$i) {
  211. $t += $this->V[$i][$k]* $this->V[$i][$j];
  212. }
  213. $t = -$t / $this->V[$k+1][$k];
  214. for ($i = $k + 1; $i < $this->n; ++$i) {
  215. $this->V[$i][$j] += $t * $this->V[$i][$k];
  216. }
  217. }
  218. }
  219. for ($i = 0; $i < $this->n; ++$i) {
  220. $this->V[$i][$k] = 0.0;
  221. }
  222. $this->V[$k][$k] = 1.0;
  223. }
  224. }
  225. // Main iteration loop for the singular values.
  226. $pp = $p - 1;
  227. $iter = 0;
  228. $eps = pow(2.0, -52.0);
  229. while ($p > 0) {
  230. // Here is where a test for too many iterations would go.
  231. // This section of the program inspects for negligible
  232. // elements in the s and e arrays. On completion the
  233. // variables kase and k are set as follows:
  234. // kase = 1 if s(p) and e[k-1] are negligible and k<p
  235. // kase = 2 if s(k) is negligible and k<p
  236. // kase = 3 if e[k-1] is negligible, k<p, and
  237. // s(k), ..., s(p) are not negligible (qr step).
  238. // kase = 4 if e(p-1) is negligible (convergence).
  239. for ($k = $p - 2; $k >= -1; --$k) {
  240. if ($k == -1) {
  241. break;
  242. }
  243. if (abs($e[$k]) <= $eps * (abs($this->s[$k]) + abs($this->s[$k+1]))) {
  244. $e[$k] = 0.0;
  245. break;
  246. }
  247. }
  248. if ($k == $p - 2) {
  249. $kase = 4;
  250. } else {
  251. for ($ks = $p - 1; $ks >= $k; --$ks) {
  252. if ($ks == $k) {
  253. break;
  254. }
  255. $t = ($ks != $p ? abs($e[$ks]) : 0.) + ($ks != $k + 1 ? abs($e[$ks-1]) : 0.);
  256. if (abs($this->s[$ks]) <= $eps * $t) {
  257. $this->s[$ks] = 0.0;
  258. break;
  259. }
  260. }
  261. if ($ks == $k) {
  262. $kase = 3;
  263. } elseif ($ks == $p-1) {
  264. $kase = 1;
  265. } else {
  266. $kase = 2;
  267. $k = $ks;
  268. }
  269. }
  270. ++$k;
  271. // Perform the task indicated by kase.
  272. switch ($kase) {
  273. // Deflate negligible s(p).
  274. case 1:
  275. $f = $e[$p-2];
  276. $e[$p-2] = 0.0;
  277. for ($j = $p - 2; $j >= $k; --$j) {
  278. $t = hypo($this->s[$j], $f);
  279. $cs = $this->s[$j] / $t;
  280. $sn = $f / $t;
  281. $this->s[$j] = $t;
  282. if ($j != $k) {
  283. $f = -$sn * $e[$j-1];
  284. $e[$j-1] = $cs * $e[$j-1];
  285. }
  286. if ($wantv) {
  287. for ($i = 0; $i < $this->n; ++$i) {
  288. $t = $cs * $this->V[$i][$j] + $sn * $this->V[$i][$p-1];
  289. $this->V[$i][$p-1] = -$sn * $this->V[$i][$j] + $cs * $this->V[$i][$p-1];
  290. $this->V[$i][$j] = $t;
  291. }
  292. }
  293. }
  294. break;
  295. // Split at negligible s(k).
  296. case 2:
  297. $f = $e[$k-1];
  298. $e[$k-1] = 0.0;
  299. for ($j = $k; $j < $p; ++$j) {
  300. $t = hypo($this->s[$j], $f);
  301. $cs = $this->s[$j] / $t;
  302. $sn = $f / $t;
  303. $this->s[$j] = $t;
  304. $f = -$sn * $e[$j];
  305. $e[$j] = $cs * $e[$j];
  306. if ($wantu) {
  307. for ($i = 0; $i < $this->m; ++$i) {
  308. $t = $cs * $this->U[$i][$j] + $sn * $this->U[$i][$k-1];
  309. $this->U[$i][$k-1] = -$sn * $this->U[$i][$j] + $cs * $this->U[$i][$k-1];
  310. $this->U[$i][$j] = $t;
  311. }
  312. }
  313. }
  314. break;
  315. // Perform one qr step.
  316. case 3:
  317. // Calculate the shift.
  318. $scale = max(max(max(max(abs($this->s[$p-1]), abs($this->s[$p-2])), abs($e[$p-2])), abs($this->s[$k])), abs($e[$k]));
  319. $sp = $this->s[$p-1] / $scale;
  320. $spm1 = $this->s[$p-2] / $scale;
  321. $epm1 = $e[$p-2] / $scale;
  322. $sk = $this->s[$k] / $scale;
  323. $ek = $e[$k] / $scale;
  324. $b = (($spm1 + $sp) * ($spm1 - $sp) + $epm1 * $epm1) / 2.0;
  325. $c = ($sp * $epm1) * ($sp * $epm1);
  326. $shift = 0.0;
  327. if (($b != 0.0) || ($c != 0.0)) {
  328. $shift = sqrt($b * $b + $c);
  329. if ($b < 0.0) {
  330. $shift = -$shift;
  331. }
  332. $shift = $c / ($b + $shift);
  333. }
  334. $f = ($sk + $sp) * ($sk - $sp) + $shift;
  335. $g = $sk * $ek;
  336. // Chase zeros.
  337. for ($j = $k; $j < $p-1; ++$j) {
  338. $t = hypo($f, $g);
  339. $cs = $f/$t;
  340. $sn = $g/$t;
  341. if ($j != $k) {
  342. $e[$j-1] = $t;
  343. }
  344. $f = $cs * $this->s[$j] + $sn * $e[$j];
  345. $e[$j] = $cs * $e[$j] - $sn * $this->s[$j];
  346. $g = $sn * $this->s[$j+1];
  347. $this->s[$j+1] = $cs * $this->s[$j+1];
  348. if ($wantv) {
  349. for ($i = 0; $i < $this->n; ++$i) {
  350. $t = $cs * $this->V[$i][$j] + $sn * $this->V[$i][$j+1];
  351. $this->V[$i][$j+1] = -$sn * $this->V[$i][$j] + $cs * $this->V[$i][$j+1];
  352. $this->V[$i][$j] = $t;
  353. }
  354. }
  355. $t = hypo($f, $g);
  356. $cs = $f/$t;
  357. $sn = $g/$t;
  358. $this->s[$j] = $t;
  359. $f = $cs * $e[$j] + $sn * $this->s[$j+1];
  360. $this->s[$j+1] = -$sn * $e[$j] + $cs * $this->s[$j+1];
  361. $g = $sn * $e[$j+1];
  362. $e[$j+1] = $cs * $e[$j+1];
  363. if ($wantu && ($j < $this->m - 1)) {
  364. for ($i = 0; $i < $this->m; ++$i) {
  365. $t = $cs * $this->U[$i][$j] + $sn * $this->U[$i][$j+1];
  366. $this->U[$i][$j+1] = -$sn * $this->U[$i][$j] + $cs * $this->U[$i][$j+1];
  367. $this->U[$i][$j] = $t;
  368. }
  369. }
  370. }
  371. $e[$p-2] = $f;
  372. $iter = $iter + 1;
  373. break;
  374. // Convergence.
  375. case 4:
  376. // Make the singular values positive.
  377. if ($this->s[$k] <= 0.0) {
  378. $this->s[$k] = ($this->s[$k] < 0.0 ? -$this->s[$k] : 0.0);
  379. if ($wantv) {
  380. for ($i = 0; $i <= $pp; ++$i) {
  381. $this->V[$i][$k] = -$this->V[$i][$k];
  382. }
  383. }
  384. }
  385. // Order the singular values.
  386. while ($k < $pp) {
  387. if ($this->s[$k] >= $this->s[$k+1]) {
  388. break;
  389. }
  390. $t = $this->s[$k];
  391. $this->s[$k] = $this->s[$k+1];
  392. $this->s[$k+1] = $t;
  393. if ($wantv and ($k < $this->n - 1)) {
  394. for ($i = 0; $i < $this->n; ++$i) {
  395. $t = $this->V[$i][$k+1];
  396. $this->V[$i][$k+1] = $this->V[$i][$k];
  397. $this->V[$i][$k] = $t;
  398. }
  399. }
  400. if ($wantu and ($k < $this->m-1)) {
  401. for ($i = 0; $i < $this->m; ++$i) {
  402. $t = $this->U[$i][$k+1];
  403. $this->U[$i][$k+1] = $this->U[$i][$k];
  404. $this->U[$i][$k] = $t;
  405. }
  406. }
  407. ++$k;
  408. }
  409. $iter = 0;
  410. --$p;
  411. break;
  412. } // end switch
  413. } // end while
  414. } // end constructor
  415. /**
  416. * Return the left singular vectors
  417. *
  418. * @access public
  419. * @return U
  420. */
  421. public function getU()
  422. {
  423. return new Matrix($this->U, $this->m, min($this->m + 1, $this->n));
  424. }
  425. /**
  426. * Return the right singular vectors
  427. *
  428. * @access public
  429. * @return V
  430. */
  431. public function getV()
  432. {
  433. return new Matrix($this->V);
  434. }
  435. /**
  436. * Return the one-dimensional array of singular values
  437. *
  438. * @access public
  439. * @return diagonal of S.
  440. */
  441. public function getSingularValues()
  442. {
  443. return $this->s;
  444. }
  445. /**
  446. * Return the diagonal matrix of singular values
  447. *
  448. * @access public
  449. * @return S
  450. */
  451. public function getS()
  452. {
  453. for ($i = 0; $i < $this->n; ++$i) {
  454. for ($j = 0; $j < $this->n; ++$j) {
  455. $S[$i][$j] = 0.0;
  456. }
  457. $S[$i][$i] = $this->s[$i];
  458. }
  459. return new Matrix($S);
  460. }
  461. /**
  462. * Two norm
  463. *
  464. * @access public
  465. * @return max(S)
  466. */
  467. public function norm2()
  468. {
  469. return $this->s[0];
  470. }
  471. /**
  472. * Two norm condition number
  473. *
  474. * @access public
  475. * @return max(S)/min(S)
  476. */
  477. public function cond()
  478. {
  479. return $this->s[0] / $this->s[min($this->m, $this->n) - 1];
  480. }
  481. /**
  482. * Effective numerical matrix rank
  483. *
  484. * @access public
  485. * @return Number of nonnegligible singular values.
  486. */
  487. public function rank()
  488. {
  489. $eps = pow(2.0, -52.0);
  490. $tol = max($this->m, $this->n) * $this->s[0] * $eps;
  491. $r = 0;
  492. for ($i = 0; $i < count($this->s); ++$i) {
  493. if ($this->s[$i] > $tol) {
  494. ++$r;
  495. }
  496. }
  497. return $r;
  498. }
  499. }