multiroc column issue

[Gster963]

I have been trying to use multi_roc for a few days and I run into this error.

I have 4 binary predictors of an outcome (DX) and would like to run multi-ROC to generate the AUC.
I get the error below

summary(data.complete)
DX_pred_febsz DX_pred_etio DX_pred_enceph DX_pred_szfreq DX_true
Min. :0.0000 Min. :0.0000 Min. :0.0000 Min. :0.0000 Min. :0.0000
1st Qu.:0.0000 1st Qu.:0.0000 1st Qu.:0.0000 1st Qu.:1.0000 1st Qu.:0.0000
Median :1.0000 Median :0.0000 Median :0.0000 Median :1.0000 Median :1.0000
Mean :0.7266 Mean :0.4865 Mean :0.2229 Mean :0.9224 Mean :0.6675
3rd Qu.:1.0000 3rd Qu.:1.0000 3rd Qu.:0.0000 3rd Qu.:1.0000 3rd Qu.:1.0000
Max. :1.0000 Max. :1.0000 Max. :1.0000 Max. :1.0000 Max. :1.0000

multiroc.data <- multi_roc(data.complete, force_diag = TRUE)

Error in colnames<-(*tmp*, value = character(0)) :
attempt to set 'colnames' on an object with less than two dimensions

[behoppy]

I am having a similar issue!

[WandeRum]

Since the multi_roc is made for multiple classes (>2), there should be multiple (>2) columns of the true labels. You can check the README to see how to generate a dummy matrix of true labels.

[graflaura]

I am getting the same error even though I have multiple columns of the true labels. I used the same code for creating dummy variables as in the example (using the dummies package). So my dataframe looks like this:

'data.frame': 205 obs. of 6 variables:
$ S1_true: int 0 1 0 0 0 0 0 0 1 0 ...
$ S2_true: int 1 0 1 0 1 1 1 1 0 1 ...
$ S3_true: int 0 0 0 1 0 0 0 0 0 0 ...
$ S1_pred: num 0.1041 0.91388 0.07417 0.50429 0.00732 ...
$ S2_pred: num 0.8886 0.0649 0.9139 0.2255 0.9908 ...
$ S3_pred: num 0.00729 0.02121 0.01194 0.27023 0.00184 ...

I tried changing the variables "S1_true" to "S3_true" to character instead of integer because the error says something about character, but that didn't help.

@WandeRum Do you have any idea what is causing the error and how to fix it?

[WandeRum]

Thanks for the bug report. Could u tell me how large is your data set and how many samples in each group? It could happen if one group gets very small number of samples. And I am glad to test your data if feasible.

[emarosti]

I get this error with a similar dataframe to @graflaura :

'data.frame':	59751 obs. of  6 variables:
 $ class0_true: num  0 0 1 1 0 0 0 1 1 0 ...
 $ class1_true: num  0 1 0 0 0 0 0 0 0 1 ...
 $ class2_true: num  1 0 0 0 1 1 1 0 0 0 ...
 $ class0_pred: num  0.492 0.399 0.652 0.413 0.544 ...
 $ class1_pred: num  0.0557 0.1663 0.0988 0.0763 0.0991 ...
 $ class2_pred: num  0.453 0.435 0.249 0.511 0.357 ...

I was able to bypass the error by changing the column names of the predicted values to have a _nn suffix:

'data.frame':	59751 obs. of  6 variables:
 $ class0_true   : num  0 0 1 1 0 0 0 1 1 0 ...
 $ class1_true   : num  0 1 0 0 0 0 0 0 0 1 ...
 $ class2_true   : num  1 0 0 0 1 1 1 0 0 0 ...
 $ class0_pred_nn: num  0.492 0.399 0.652 0.413 0.544 ...
 $ class1_pred_nn: num  0.0557 0.1663 0.0988 0.0763 0.0991 ...
 $ class2_pred_nn: num  0.453 0.435 0.249 0.511 0.357 ...

Therefore, by only changing the predicted column names to have some type of suffix after _pred, the error was avoided.

For reference, here's the full traceback @WandeRum :

Error in `colnames<-`(`*tmp*`, value = character(0)) : attempt to set 'colnames' on an object with less than two dimensions

6. stop("attempt to set 'colnames' on an object with less than two dimensions")
5.`colnames<-`(`*tmp*`, value = character(0))
4. eval(call("<-", lhs, result[["value"]]), parent, parent)
3. eval(call("<-", lhs, result[["value"]]), parent, parent) 
2. colnames(y_pred) %<>% gsub("_pred.*", "", .)
1. multi_roc(comb, force_diag = T)

Thanks for creating this R package!

[WandeRum]

I'd like to test your data and fix the bug if you can share either through GitHub or my personal email: wander1021@gmail.com. Thank you.