0

我正在 R 中尝试一个 subervised SOM，它适用于 R 版本 3.1.0 但不适用于 3.3.2。这是它在 3.1.0 中正常工作：

> fitControl <- trainControl(method = "repeatedcv",number = 10,repeats = 10)
> somFit1 <- train(Species ~ ., data = iris, method = "xyf",preProc = c("center","scale"),trControl = fitControl,metric="Kappa")
Loading required package: kohonen
Loading required package: class
Loading required package: MASS
Warning messages:
1: package ‘kohonen’ was built under R version 3.1.3 
2: package ‘class’ was built under R version 3.1.3 
3: package ‘MASS’ was built under R version 3.1.3 
> somFit1
Self-Organizing Maps 

150 samples
  4 predictor
  3 classes: 'setosa', 'versicolor', 'virginica' 

Pre-processing: centered (4), scaled (4) 
Resampling: Cross-Validated (10 fold, repeated 10 times) 
Summary of sample sizes: 135, 135, 135, 135, 135, 135, ... 
Resampling results across tuning parameters:

  xdim  ydim  xweight  Accuracy   Kappa  Accuracy SD  Kappa SD  
  1     2     0.5      0.6666667  0.500  0.00000000   0.00000000
  1     2     0.7      0.6666667  0.500  0.00000000   0.00000000
  1     2     0.9      0.6666667  0.500  0.00000000   0.00000000
  1     3     0.5      0.8040000  0.706  0.12128012   0.18192017
  1     3     0.7      0.8140000  0.721  0.11996820   0.17995229
  1     3     0.9      0.8240000  0.736  0.09309252   0.13963878
  1     4     0.5      0.8806667  0.821  0.08384423   0.12576634    
  1     4     0.7      0.8953333  0.843  0.07043555   0.10565333
  1     4     0.9      0.8386667  0.758  0.08693819   0.13040729
  2     2     0.5      0.8960000  0.844  0.08649296   0.12973944
  2     2     0.7      0.8833333  0.825  0.09057740   0.13586610
  2     2     0.9      0.8313333  0.747  0.08499290   0.12748935
  2     3     0.5      0.8793333  0.819  0.07796226   0.11694339
  2     3     0.7      0.9100000  0.865  0.07602646   0.11403969
  2     3     0.9      0.9326667  0.899  0.06930471   0.10395706
  2     4     0.5      0.9106667  0.866  0.06977276   0.10465914
  2     4     0.7      0.9193333  0.879  0.07478814   0.11218221
  2     4     0.9      0.9453333  0.918  0.06021287   0.09031930
  3     3     0.5      0.9333333  0.900  0.05763775   0.08645662
  3     3     0.7      0.9353333  0.903  0.06317808   0.09476713
  3     3     0.9      0.9440000  0.916  0.05336699   0.08005049
  3     4     0.5      0.9406667  0.911  0.06205967   0.09308951
  3     4     0.7      0.9366667  0.905  0.06658244   0.09987366
  3     4     0.9      0.9480000  0.922  0.05808775   0.08713162

Tuning parameter 'topo' was held constant at a value of hexagonal
Kappa was used to select the optimal model using  the largest value.
The final values used for the model were xdim = 3, ydim = 4, xweight = 0.9
 and topo = hexagonal.

在这里它在 3.3.2 中不起作用：

> fitControl <- trainControl(method = "repeatedcv",number = 10,repeats = 10)
> somFit1 <- train(Species ~ ., data = iris, method = "xyf",preProc = c("center","scale"),trControl = fitControl,metric="Kappa")  
Loading required package: kohonen
Something is wrong; all the Kappa metric values are missing:
    Accuracy       Kappa    
 Min.   : NA   Min.   : NA  
 1st Qu.: NA   1st Qu.: NA  
 Median : NA   Median : NA  
 Mean   :NaN   Mean   :NaN  
 3rd Qu.: NA   3rd Qu.: NA  
 Max.   : NA   Max.   : NA  
  NA's   :24    NA's   :24   
Error in train.default(x, y, weights = w, ...) : Stopping
In addition: There were 50 or more warnings (use warnings() to see the first 50)

警告是：

Warning messages:
1: package ‘kohonen’ was built under R version 3.3.3
2: In eval(expr, envir, enclos) :
  model fit failed for Fold01.Rep01: xdim=1, ydim=2, xweight=0.5, topo=hexagonal Error in supersom(list(X, Y), ...) : 
  unused arguments (xweight = 0.5, contin = FALSE)

3: In eval(expr, envir, enclos) :
  model fit failed for Fold01.Rep01: xdim=2, ydim=2, xweight=0.5, topo=hexagonal Error in supersom(list(X, Y), ...) : 
  unused arguments (xweight = 0.5, contin = FALSE)

我觉得未使用的论点很重要。在过去使用插入符号时，当我在对插入符号的调用中删除多余的“verbose = FALSE”参数时，该错误消息已经消失。但是在这里我不知道如何阻止 xweight 和 contin 通过。谢谢。

4

1 回答 1

0

我现在才看到这个，我在插入符号的 SOM 上工作。但似乎你没有包括 user.weights。首先，我使用的是 Windows ;-( 但是嘿嘿。使用并行处理来加快速度：

install.packages("doParallel")
library(doParallel)   # https://github.com/tobigithub/R-parallel/wiki/
library(foreach)
#Register cores -1 to allow one core for other work
 cl <- makeCluster(detectCores()-1)
 registerDoParallel(cl)
 getDoParWorkers()

现在我有一个预处理、训练控制和一个调谐网格。然后我终于训练并跟踪训练 SOM 所需的时间

C_preProcess.method <- c("zv", "nzv", "corr", "center", "scale")
fitControl2 <- trainControl(method = "repeatedcv", number = 10, repeats = 5, savePredictions = 'final', classProbs = TRUE, summaryFunction=twoClassSummary)
tg5 <- expand.grid(xdim = c(20,10), ydim=c(20,10), user.weights=c(1,3), topo="hexagonal") #seq(from = 1, to = 10, by = 1))

t5<-system.time(Fit5 <- train(shares ~ ., data = caret.train.data, method = 'xyf', preProcess= C_preProcess.method, metric = "ROC", trControl = fitControl2, tuneGrid=tg5))

我希望这可以帮助那里的人...

于 2018-04-02T12:55:12.367 回答