Question

我正在寻找时间序列的SSE功能。

当使用Kmedoid函数时，例如来自pam包的cluster，返回类没有包含来自聚类的SSE值的索引。话虽这么说，我做了一个简单的函数，使用距离矩阵（已经是pam函数的参数），但它并不那么快，考虑到一个大约4亿元素的巨大矩阵，是一个输入。

出于这个原因，是否有任何已经具有SSE功能的包，对于这种情况？或者，我该如何改进我的功能？输入的示例可以是：

( ts dataset or the d.matrix ,the clustering classes )

我有谷歌：SSE R功能在找到包时没有成功。

所以，到目前为止我所做的功能就是这个：

generate.sse <- function( dmatrix, groups )
{
  sum = 0
  for(i in 1:nrow(dmatrix))
  {
    sum = sum + sum(dmatrix[ i , groups == groups[i] ])
  }
  return (sum)
}

正如我之前所说，它不是最佳的，因此它很慢。主要是因为它的用法和总和，因为它是C之类的。

对于一个可重现的例子，我将留下一个时间序列的例子（dmatrix可以用cor函数获得）和一个可能从pam函数返回的聚类组。两者都是使用dput

生成的

时间序列：

structure(c(0, 311, 0, 739, 277, 841, 0, 548, 177, 36, 0, 0, 
0, 0, 1268, 30, 547, 0, 95, 0, 2, 0, 0, 0, 2344, 0, 8, 574, 0, 
0, 0, 589, 0, 0, 40, 3, 0, 2, 0, 0, 28, 704, 0, 46, 0, 1, 0, 
52, 1374, 0, 2298, 0, 2, 0, 27, 827, 12, 0, 19, 0, 10, 815, 0, 
106, 0, 0, 0, 852, 1113, 0, 79, 12, 0, 0, 0, 914, 0, 0, 0, 69, 
0, 0, 78, 0, 51, 0, 1841, 0, 314, 1047, 0, 0, 0, 1522, 2251, 
0, 55, 0, 0, 0, 20, 1338, 762, 1462, 972, 1877, 3, 1717, 743, 
4, 0, 29, 0, 658, 5436, 65, 981, 11, 1866, 20, 147, 29, 134, 
23, 7241, 144, 0, 553, 152, 827, 300, 80, 252, 58, 159, 274, 
2687, 69, 4198, 115, 9, 5046, 0, 4, 60, 122, 20, 309, 58, 377, 
7045, 0, 996, 387, 311, 4911, 8, 22, 124, 119, 324, 2260, 22, 
1062, 0, 10, 28, 2379, 5545, 370, 222, 178, 208, 189, 24, 2917, 
851, 1012, 43, 128, 1756, 484, 0, 1275, 144, 630, 7120, 698, 
0, 3, 0, 787, 24, 7305, 6495, 584, 476, 226, 332, 15, 18, 1087, 
1022, 1946, 1029, 2120, 16, 1623, 678, 29, 0, 31, 0, 604, 5590, 
50, 950, 0, 1758, 17, 141, 27, 1, 22, 7846, 181, 0, 1637, 41, 
942, 332, 51, 211, 38, 138, 261, 2379, 69, 4297, 113, 36, 4836, 
3, 231, 73, 84, 24, 357, 56, 367, 7920, 2, 1114, 250, 228, 4509, 
6, 19, 181, 68, 759, 2048, 21, 959, 0, 12, 38, 2388, 5501, 385, 
264, 272, 145, 90, 94, 2974, 948, 1260, 42, 204, 1688, 297, 0, 
1145, 221, 24, 8042, 442, 0, 0, 0, 837, 17, 6015, 6649, 518, 
534, 523, 314, 26, 18, 1420, 1517, 1570, 986, 2665, 17, 1587, 
683, 23, 16, 23, 0, 500, 5814, 10, 992, 0, 2185, 23, 120, 19, 
17, 28, 4043, 102, 0, 1872, 110, 852, 256, 66, 265, 40, 132, 
307, 2575, 69, 1, 63, 44, 5734, 0, 120, 124, 27, 17, 364, 66, 
545, 8229, 1, 970, 212, 164, 4683, 11, 22, 68, 49, 414, 1028, 
23, 1060, 0, 18, 21, 2234, 4775, 387, 242, 212, 158, 118, 166, 
1118, 782, 1318, 122, 422, 2153, 341, 61, 1094, 164, 566, 7264, 
515, 0, 0, 0, 811, 21, 5946, 6675, 528, 700, 447, 328, 25, 10, 
1319, 1530, 1564, 966, 2085, 13, 1296, 695, 45, 17, 26, 76, 504, 
6335, 9, 928, 0, 2248, 26, 165, 0, 40, 15, 9348, 140, 0, 1904, 
183, 0, 115, 13, 254, 0, 180, 253, 3524, 64, 4043, 97, 18, 4505, 
0, 235, 69, 42, 18, 647, 516, 477, 8544, 3, 894, 256, 299, 4873, 
7, 20, 117, 31, 321, 1897, 23, 1606, 0, 16, 32, 2451, 5031, 419, 
258, 149, 146, 71, 155, 2860, 622, 0, 0, 142, 1677, 459, 0, 943, 
153, 20, 6869, 603, 0, 0, 21, 0, 21, 5545, 7052, 583, 491, 545, 
359, 20, 10, 1257, 1571, 1542, 1011, 2300, 17, 1646, 58, 53, 
0, 19, 0, 565, 5478, 23, 638, 0, 1941, 16, 119, 11, 1, 20, 7018, 
119, 0, 1795, 89, 0, 299, 57, 200, 71, 149, 262, 2326, 45, 4942, 
0, 11, 4994, 0, 91, 75, 53, 15, 252, 448, 323, 6923, 1, 1101, 
314, 338, 4656, 12, 15, 102, 43, 283, 2496, 21, 1282, 24, 16, 
24, 2245, 7817, 365, 222, 255, 155, 66, 96, 2866, 738, 2783, 
198, 256, 1521, 311, 0, 925, 203, 465, 6728, 483, 0, 0, 0, 692, 
6, 5188, 6343, 598, 507, 271, 297, 18, 0, 679, 840, 1123, 450, 
1597, 32, 856, 199, 43, 0, 0, 0, 601, 3544, 0, 553, 0, 1010, 
1, 154, 0, 20, 0, 3672, 0, 20, 1140, 54, 540, 0, 18, 166, 0, 
188, 43, 1685, 29, 9, 0, 49, 1777, 0, 23, 18, 15, 0, 56, 80, 
0, 3439, 0, 701, 107, 161, 1756, 5, 0, 0, 167, 30, 1426, 0, 309, 
0, 12, 0, 1470, 1829, 120, 105, 34, 0, 23, 73, 1376, 0, 0, 0, 
274, 748, 86, 164, 289, 83, 0, 3726, 318, 15, 0, 17, 165, 0, 
2819, 2882, 39, 153, 170, 0, 0), .Dim = c(100L, 7L))

组：

c(1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 2L, 1L, 1L, 1L, 1L, 1L, 
1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 2L, 1L, 1L, 1L, 1L, 
2L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 2L, 1L, 1L, 1L, 1L, 1L, 1L, 
1L, 2L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 2L, 1L, 1L, 1L, 1L, 1L, 1L, 
1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 
1L, 1L, 1L, 2L, 1L, 1L, 1L, 1L, 1L, 2L, 2L, 2L, 1L, 1L, 1L, 1L, 
1L, 1L, 1L, 1L, 1L)

Answer 1

PAM不使用也不优化SSE。

相反，k-medoids的目标是最小化总偏差 TD。我假设pam函数在返回的对象中包含了这个分数。

R本身慢。如果您有任何更大的计算，请避免使用R代码，而是重写代码以尽可能使用C和FORTRAN。也许你可以使用一些“矢量化”操作，或者你可以在C中编写自己的库。考虑R只是一个“驱动”语言，就像底层库的“用户界面”。

具有SSE功能的包用于庞大的数据集 - 或 - 实现更快的功能

1 个答案: