Carry out the manyROC analysis

Source: R/print-methods.R, R/roc_manyroc.R

Do the ROC analysis (roc_analysis) for each column of x.

# S3 method for manyroc_result
print(x, ..., show_all = FALSE, perf_digits = 2, fmt = "%.3g")

roc_manyroc(x, gr = NULL, optimize_by = "bac", ...)

# S3 method for matrix
roc_manyroc(x, gr = NULL, optimize_by = "bac", ..., gr_sep = " vs. ")

# S3 method for numeric
roc_manyroc(x, gr = NULL, optimize_by = "bac", ...)

# S3 method for data.frame
roc_manyroc(x, gr = NULL, optimize_by = "bac", ...)

# S3 method for hyperSpec
roc_manyroc(x, gr = NULL, optimize_by = "bac", ...)

Arguments

x

A numeric matrix, a data frame, a hyperSpec object or another type of object, convertible to a numeric matrix.
...

Arguments passed for further methods.
show_all

(logical(1))
A flag if whole dataset should be printed. If FALSE, only a few fisrt and last rows will be printed.
perf_digits

(integer(1))
A number of decimals to display for performance measures. Default is 2.
fmt

(character(1))
A string indicating number display format for other numeric columns excluding performance measures. The string will be passed to sprintf. Default is "%.3g".
gr

Either a string (scalar, character(1)) or a factor variable (a vector) which defines groups in x.
optimize_by

(string(1))
A string with the name of classification performance measure to use. Currently available options:

  • "bac" - for balanced accuracy (mean of sensitivity and specificity);

  • "kappa" - for Cohens kappa;

  • "youden" - for Youden's index;
gr_sep

Group separator used to paste the names of groups. Default is " vs. ".

Value

Object of classes manyroc_result and data.frame with columns:

  • compared_groups Names of compared groups (separated by gr_sep with default value " vs. ");

  • feature names of numeric features used in analysis;

  • median_neg median value of negatives group;

  • cutoff for optimal threshold/cut-off values and corresponding performance measures;

  • median_pos median value positives group;

  • TP number of true positives;

  • FN number of false negatives;

  • FP number of false positives;

  • TN number of true negatives;

  • sens sensitivity (true positive rate, recall);

  • spec specificity (true negative rate);

  • PPV positive predictive value (precision);

  • NPV negative predictive value;

  • BAC balanced accuracy;

  • Youden Youden’s J index;

  • Kappa Cohen's kappa;

  • AUC area under the ROC curve;

Details

  1. Consider ordering of factor gr levels before the analysis, as the first level will always be treated as negative and the last as positive. E.g., if we have factor with 3 levels in this particular order "A", "B", "C", then "A" will always be negative, "C" always positive and "B" positive, when compared to "A" and negative, when compared to "C". The same principle applies if there are more than 3 levels.
    This is important determining what specificity and sensitivity, etc., means in the context of group names: if positive is "A" then sensitivity will be related to group "A", but if "A" is negative, then specificity will be related to this group, and sensitivity to the other group.

  2. For spectroscopic data: compare spectra of each pair of indicated groups at each wavelength.

Author

Vilmantas Gegzna

Examples


library(manyROC)


# --- For numeric vectors objects ---

data(PlantGrowth)
roc_manyroc(x = PlantGrowth$weight, gr = PlantGrowth$group)

#>   compared_groups feature median_neg cutoff median_pos tp fn fp tn sens spec
#> 1   ctrl vs. trt1       1       5.15   4.37       4.55  5  5  1  9 0.50 0.90
#> 2   ctrl vs. trt2       1       5.15   5.22       5.44  8  2  3  7 0.80 0.70
#> 3   trt1 vs. trt2       1       4.55    4.9       5.44 10  0  2  8 1.00 0.80
#>    ppv  npv  bac youden kappa  auc
#> 1 0.83 0.64 0.70   0.40  0.40 0.67
#> 2 0.73 0.78 0.75   0.50  0.50 0.75
#> 3 0.83 1.00 0.90   0.80  0.80 0.84


# --- For dataframes objects ---

data(CO2)
roc_manyroc(CO2[, c("conc", "uptake")], CO2$Type)

#>          compared_groups feature median_neg cutoff median_pos tp fn fp tn sens
#> 1 Quebec vs. Mississippi    conc        350    200        350 18 24 18 24 0.43
#> 2 Quebec vs. Mississippi  uptake       37.2   32.3       19.3 41  1 13 29 0.98
#>   spec  ppv  npv  bac youden kappa  auc
#> 1 0.57 0.50 0.50 0.50   0.00  0.00 0.50
#> 2 0.69 0.76 0.97 0.83   0.67  0.67 0.84

data(OrchardSprays)
roc_manyroc(OrchardSprays$decrease, OrchardSprays$treatment)

#>     compared_groups feature median_neg cutoff median_pos  tp  fn  fp  tn sens
#> 1           A vs. B       1          4    5.5        7.5   6   2   1   7 0.75
#> 2           A vs. C       1          4   12.5       16.5   7   1   0   8 0.88
#> 3           A vs. D       1          4     16         32   8   0   0   8 1.00
#> 4           A vs. E       1          4   25.5         53   8   0   0   8 1.00
#> ...             ...     ...        ...    ...        ... ... ... ... ...  ...
#> 25          E vs. H       1         53     65         81   8   0   2   6 1.00
#> 26          F vs. G       1         70     70         72   6   2   4   4 0.75
#> 27          F vs. H       1         70   71.5         81   7   1   3   5 0.88
#> 28          G vs. H       1         72   80.5         81   5   3   1   7 0.62
#>                                                                              
#>     spec  ppv  npv  bac youden kappa  auc
#> 1   0.88 0.86 0.78 0.81   0.62  0.62 0.80
#> 2   1.00 1.00 0.89 0.94   0.88  0.88 0.98
#> 3   1.00 1.00 1.00 1.00   1.00  1.00 1.00
#> 4   1.00 1.00 1.00 1.00   1.00  1.00 1.00
#> ...  ...  ...  ...  ...    ...   ...  ...
#> 25  0.75 0.80 1.00 0.88   0.75  0.75 0.81
#> 26  0.50 0.60 0.67 0.62   0.25  0.25 0.55
#> 27  0.62 0.70 0.83 0.75   0.50  0.50 0.70
#> 28  0.88 0.83 0.70 0.75   0.50  0.50 0.75
#>                                          

# --- For hyperSpec objects ---

library(hyperSpec)
fluorescence

#> hyperSpec object
#>    150 spectra
#>    3 data columns
#>    501 data points / spectrum
#> wavelength: lambda/nm [integer] 300 301 ... 800 
#> data:  (150 rows x 3 columns)
#>    1. gr:  [factor] B B ... A 
#>    2. class:  [factor] N l ... S1 
#>    3. spc: I / a.u. [matrix, array501] 159.8996 139.9296 ... 12.11558 

roc_manyroc(fluorescence[, , 500 ~ 502], fluorescence$gr)

#>   compared_groups feature median_neg cutoff median_pos tp fn fp tn sens spec
#> 1         A vs. B       1        338    335        333 35 20 24 28 0.64 0.54
#> 2         A vs. B       2        335    328        327 29 26 19 33 0.53 0.63
#> 3         A vs. B       3        335    332        329 36 19 24 28 0.65 0.54
#> 4         A vs. C       1        338    329        334 21 22 20 32 0.49 0.62
#> 5         A vs. C       2        335    329        334 21 22 20 32 0.49 0.62
#> 6         A vs. C       3        335    340        340 22 21 22 30 0.51 0.58
#> 7         B vs. C       1        333    343        334 21 22 15 40 0.49 0.73
#> 8         B vs. C       2        327    344        334 21 22 15 40 0.49 0.73
#> 9         B vs. C       3        329    339        340 22 21 16 39 0.51 0.71
#>    ppv  npv  bac youden kappa  auc
#> 1 0.59 0.58 0.59   0.17  0.18 0.57
#> 2 0.60 0.56 0.58   0.16  0.16 0.56
#> 3 0.60 0.60 0.60   0.19  0.19 0.56
#> 4 0.51 0.59 0.55   0.10  0.10 0.50
#> 5 0.51 0.59 0.55   0.10  0.10 0.49
#> 6 0.50 0.59 0.54   0.09  0.09 0.51
#> 7 0.58 0.65 0.61   0.22  0.22 0.57
#> 8 0.58 0.65 0.61   0.22  0.22 0.57
#> 9 0.58 0.65 0.61   0.22  0.22 0.56