将参数1定义为1减去参数2，使用R的paradox包。

huangapple
go评论101阅读模式
英文:

Define parameter1 as 1 - parameter2 using R paradox package

问题

我想要使用paradox包将parameter1定义为1 - parameter2。这意味着parameter1依赖于parameter2（我认为depends参数在这里不起作用）。

这是我的研究空间：

  1. search_space = ps(
  2.   # 数据预处理
  3.   interaction_branch.selection = p_fct(levels = c("nop_filter", "modelmatrix")),
  4.   winsorizesimple.probs_high = p_fct(levels = c("0.99", "0.98", "0.97")),
  5.   winsorizesimple.probs_low = p_dbl(lower = 0, upper = 1),
  6.   # ranger
  7.   ranger.ranger.max.depth = p_fct(levels = c(2L, 10L)),
  8.   ranger.ranger.splitrule = p_fct(levels = c("gini", "extratrees")),
  9.   ranger.ranger.mtry.ratio = p_dbl(0.5, 1),
  10.   # kknn
  11.   kknn.kknn.k = p_int(1, 10),
  12.   # 额外的转换
  13.   .extra_trafo = function(x, param_set) {
  14.     winsorizesimple.probs_high = switch(
  15.       x$winsorizesimple.probs_high,
  16.       "0.99" = 0.99,
  17.       "0.98" = 0.98,
  18.       "0.97" = 0.97
  19.     )
  20.     x$winsorizesimple.probs_low = 1 - winsorizesimple.probs_high
  21.     x
  22.   }
  23. )

输出如下：

  1.       interaction_branch.selection winsorizesimple.probs_high winsorizesimple.probs_low ranger.ranger.max.depth ranger.ranger.splitrule ranger.ranger.mtry.ratio kknn.kknn.k
  2.    1:                   nop_filter                       0.99                   0.01                       2                    gini                0.5000000           1
  3.    2:                   nop_filter                       0.99                   0.01                       2                    gini                0.5000000           4
  4.    3:                   nop_filter                       0.99                   0.01                       2                    gini                0.5000000           7
  5.    4:                   nop_filter                       0.99                   0.01                       2                    gini                0.5000000          10
  6.    5:                   nop_filter                       0.99                   0.01                       2                    gini                0.6666667           1
  7.   ---                                                                                                                                           
  8. 1532:                  modelmatrix                       0.97                   0.03                      10              extratrees                0.8333333          10
  9. 1533:                  modelmatrix                       0.97                   0.03                      10              extratrees                1.0000000           1
  10. 1534:                  modelmatrix                       0.97                   0.03                      10              extratrees                1.0000000           4
  11. 1535:                  modelmatrix                       0.97                   0.03                      10              extratrees                1.0000000           7
  12. 1536:                  modelmatrix                       0.97                   0.03                      10              extratrees                1.0000000          10

我理解你的需求是将winsorizesimple.probs_low从-1更改为0.01，因此这是更正后的输出。希望这对你有帮助，不需要重新生成示例。

英文:

I want to define parameter1 as 1 - parameter2 using paradox package.
That is parameter 1 depends on parameter 2 (depends argument doesn't help here I suppose).

Here is my reserach space:

  1. search_space = ps(
  2.   # preprocessing
  3.   interaction_branch.selection = p_fct(levels = c("nop_filter", "modelmatrix")),
  4.   winsorizesimple.probs_high = p_fct(levels = c("0.99", "0.98", "0.97")),
  5.   winsorizesimple.probs_low = p_dbl(lower = 0, upper = 1),
  6.   # ranger
  7.   ranger.ranger.max.depth = p_fct(levels = c(2L, 10L)),
  8.   ranger.ranger.splitrule = p_fct(levels = c("gini", "extratrees")),
  9.   ranger.ranger.mtry.ratio = p_dbl(0.5, 1),
  10.   # kknn
  11.   kknn.kknn.k = p_int(1, 10),
  12.   # extra transformations
  13.   .extra_trafo = function(x, param_set) {
  14.     winsorizesimple.probs_high = switch(
  15.       x$winsorizesimple.probs_high,
  16.       "0.99" = 0.99,
  17.       "0.98" = 0.98,
  18.       "0.97" = 0.97
  19.     )
  20.     x$winsorizesimple.probs_low = 1 - winsorizesimple.probs_high
  21.     x
  22.   }
  23. )

The ooutput is:

  1.       interaction_branch.selection winsorizesimple.probs_high winsorizesimple.probs_low ranger.ranger.max.depth ranger.ranger.splitrule ranger.ranger.mtry.ratio
  2.    1:                   nop_filter                       0.99                        -1                       2                    gini                0.5000000
  3.    2:                   nop_filter                       0.99                        -1                       2                    gini                0.5000000
  4.    3:                   nop_filter                       0.99                        -1                       2                    gini                0.5000000
  5.    4:                   nop_filter                       0.99                        -1                       2                    gini                0.5000000
  6.    5:                   nop_filter                       0.99                        -1                       2                    gini                0.6666667
  7.   ---                                                                                                                                                           
  8. 1532:                  modelmatrix                       0.97                         2                      10              extratrees                0.8333333
  9. 1533:                  modelmatrix                       0.97                         2                      10              extratrees                1.0000000
  10. 1534:                  modelmatrix                       0.97                         2                      10              extratrees                1.0000000
  11. 1535:                  modelmatrix                       0.97                         2                      10              extratrees                1.0000000
  12. 1536:                  modelmatrix                       0.97                         2                      10              extratrees                1.0000000
  13.       kknn.kknn.k
  14.    1:           1
  15.    2:           4
  16.    3:           7
  17.    4:          10
  18.    5:           1
  19.   ---            
  20. 1532:          10
  21. 1533:           1
  22. 1534:           4
  23. 1535:           7
  24. 1536:          10

I winsorizesimple.probs_high1 is 0.97, winsorizesimple.probs_low should be 0.03.
Hope you can help without reprex.

答案1

得分: 3

生成设计时出现了问题，因为对我来说，您的代码可以正常工作。

  1. library(paradox)
  2. library(data.table)
  4. search_space = ps(
  5.   # preprocessing
  6.   interaction_branch.selection = p_fct(levels = c("nop_filter", "modelmatrix")),
  7.   winsorizesimple.probs_high = p_fct(levels = c("0.99", "0.98", "0.97")),
  8.   winsorizesimple.probs_low = p_dbl(lower = 0, upper = 1),
  9.   # ranger
  10.   ranger.ranger.max.depth = p_fct(levels = c(2L, 10L)),
  11.   ranger.ranger.splitrule = p_fct(levels = c("gini", "extratrees")),
  12.   ranger.ranger.mtry.ratio = p_dbl(0.5, 1),
  13.   # kknn
  14.   kknn.kknn.k = p_int(1, 10),
  15.   # extra transformations
  16.   .extra_trafo = function(x, param_set) {
  17.     winsorizesimple.probs_high = switch(
  18.       x$winsorizesimple.probs_high,
  19.       "0.99" = 0.99,
  20.       "0.98" = 0.98,
  21.       "0.97" = 0.97
  22.     )
  23.     x$winsorizesimple.probs_low = 1 - winsorizesimple.probs_high
  24.     x
  25.   }
  26. )
  28. design = rbindlist(generate_design_grid(search_space, 3)$transpose(), fill = TRUE)
  29. design
  30. #>      interaction_branch.selection winsorizesimple.probs_high
  31. #>   1:                   nop_filter                       0.99
  32. #>   2:                   nop_filter                       0.99
  33. #>   3:                   nop_filter                       0.99
  34. #>   4:                   nop_filter                       0.99
  35. #>   5:                   nop_filter                       0.99
  36. #>  ---
  37. #> 644:                  modelmatrix                       0.97
  38. #> 645:                  modelmatrix                       0.97
  39. #> 646:                  modelmatrix                       0.97
  40. #> 647:                  modelmatrix                       0.97
  41. #> 648:                  modelmatrix                       0.97
  42. #>      winsorizesimple.probs_low ranger.ranger.max.depth ranger.ranger.splitrule
  43. #>   1:                      0.01                       2                    gini
  44. #>   2:                      0.01                       2                    gini
  45. #>   3:                      0.01                       2                    gini
  46. #>   4:                      0.01                       2                    gini
  47. #>   5:                      0.01                       2                    gini
  48. #>  ---
  49. #> 644:                      0.03                      10              extratrees
  50. #> 645:                      0.03                      10              extratrees
  51. #> 646:                      0.03                      10              extratrees
  52. #> 647:                      0.03                      10              extratrees
  53. #> 648:                      0.03                      10              extratrees
  54. #>      ranger.ranger.mtry.ratio kknn.kknn.k
  55. #>   1:                     0.50           1
  56. #>   2:                     0.50           5
  57. #>   3:                     0.50          10
  58. #>   4:                     0.75           1
  59. #>   5:                     0.75           5
  60. #>  ---
  61. #> 644:                     0.75           5
  62. #> 645:                     0.75          10
  63. #> 646:                     1.00           1
  64. #> 647:                     1.00           5
  65. #> 648:                     1.00          10

2023年2月23日创建，使用reprex v2.0.2

英文:

Something went wrong when generating the design, because for me your code works.

  1. library(paradox)
  2. library(data.table)
  4. search_space = ps(
  5.   # preprocessing
  6.   interaction_branch.selection = p_fct(levels = c("nop_filter", "modelmatrix")),
  7.   winsorizesimple.probs_high = p_fct(levels = c("0.99", "0.98", "0.97")),
  8.   winsorizesimple.probs_low = p_dbl(lower = 0, upper = 1),
  9.   # ranger
  10.   ranger.ranger.max.depth = p_fct(levels = c(2L, 10L)),
  11.   ranger.ranger.splitrule = p_fct(levels = c("gini", "extratrees")),
  12.   ranger.ranger.mtry.ratio = p_dbl(0.5, 1),
  13.   # kknn
  14.   kknn.kknn.k = p_int(1, 10),
  15.   # extra transformations
  16.   .extra_trafo = function(x, param_set) {
  17.     winsorizesimple.probs_high = switch(
  18.       x$winsorizesimple.probs_high,
  19.       "0.99" = 0.99,
  20.       "0.98" = 0.98,
  21.       "0.97" = 0.97
  22.     )
  23.     x$winsorizesimple.probs_low = 1 - winsorizesimple.probs_high
  24.     x
  25.   }
  26. )
  28. design = rbindlist(generate_design_grid(search_space, 3)$transpose(), fill = TRUE)
  29. design
  30. #>      interaction_branch.selection winsorizesimple.probs_high
  31. #>   1:                   nop_filter                       0.99
  32. #>   2:                   nop_filter                       0.99
  33. #>   3:                   nop_filter                       0.99
  34. #>   4:                   nop_filter                       0.99
  35. #>   5:                   nop_filter                       0.99
  36. #>  ---                                                        
  37. #> 644:                  modelmatrix                       0.97
  38. #> 645:                  modelmatrix                       0.97
  39. #> 646:                  modelmatrix                       0.97
  40. #> 647:                  modelmatrix                       0.97
  41. #> 648:                  modelmatrix                       0.97
  42. #>      winsorizesimple.probs_low ranger.ranger.max.depth ranger.ranger.splitrule
  43. #>   1:                      0.01                       2                    gini
  44. #>   2:                      0.01                       2                    gini
  45. #>   3:                      0.01                       2                    gini
  46. #>   4:                      0.01                       2                    gini
  47. #>   5:                      0.01                       2                    gini
  48. #>  ---                                                                          
  49. #> 644:                      0.03                      10              extratrees
  50. #> 645:                      0.03                      10              extratrees
  51. #> 646:                      0.03                      10              extratrees
  52. #> 647:                      0.03                      10              extratrees
  53. #> 648:                      0.03                      10              extratrees
  54. #>      ranger.ranger.mtry.ratio kknn.kknn.k
  55. #>   1:                     0.50           1
  56. #>   2:                     0.50           5
  57. #>   3:                     0.50          10
  58. #>   4:                     0.75           1
  59. #>   5:                     0.75           5
  60. #>  ---                                     
  61. #> 644:                     0.75           5
  62. #> 645:                     0.75          10
  63. #> 646:                     1.00           1
  64. #> 647:                     1.00           5
  65. #> 648:                     1.00          10

<sup>Created on 2023-02-23 with reprex v2.0.2</sup>

huangapple
  • 本文由 发表于 2023年2月23日 21:16:54
  • 转载请务必保留本文链接：https://go.coder-hub.com/75545365.html
  • mlr3
  • r
匿名

发表评论

匿名网友

:?: :razz: :sad: :evil: :!: :smile: :oops: :grin: :eek: :shock: :???: :cool: :lol: :mad: :twisted: :roll: :wink: :idea: :arrow: :neutral: :cry: :mrgreen:

确定