Utilities {RANKS} | R Documentation |
Utility functions
Description
Mixed utility functions to compute accuracy, norms, labels from scores and to perform stratified cross-validation.
Usage
compute.acc(pred, labels)
compute.F(pred, labels)
norm1(x)
Unit.sphere.norm(K)
do.stratified.cv.data(examples, positives, k = 5, seed = NULL)
do.cv.data(examples, positives, k = 5, seed = NULL)
labelsfromscores(scores, thresh)
Multiple.labels.from.scores(S, thresh.vect)
selection.test(pos.scores, av.scores, ind.positives, alpha = 0.05, thresh.pos = 0)
Arguments
pred |
vector of the predicted labels |
labels |
vector of the true labels. Note that 0 stands for negative and 1 for positive. In general the first level is negative and the second positive |
x |
numeric vector |
K |
a kernel matrix |
examples |
indices of the examples (a vector of integer) |
positives |
vector of integer. Indices of the positive examples. The indices refer to the indices of examples |
k |
number of folds (def = 5) |
seed |
seed of the random generator (def=NULL). If is set to NULL no initiazitation is performed |
scores |
numeric. Vector of scores: each element correspond to the score of an example |
thresh |
real value. Threshold for the classification |
S |
numeric matrix. Matrix of scores: rows represent examples, columns classes |
thresh.vect |
numeric vector. Vector of the thresholds for multiple classes (one threshold for each class) |
pos.scores |
vector with scores of positive examples. It is returned from multiple.ker.score.cv. |
av.scores |
a vector with the average scores computed by multiple.ker.score.cv. It may be a named vector. If not, the names attributes corresponding to the indices of the vector are added. |
ind.positives |
indices of the positive examples. They are the indices of av.scores corresponding to positive examples. |
alpha |
quantile level (def. 0.05) |
thresh.pos |
only values larger than thresh.pos are retained in pos.scores (def.: 0) |
Details
compute.acc
computes the accuracy for a single class
compute.F
computes the F-score for a single class
norm1
computes the L1-norm of a numeric vector
Unit.sphere.norm
normalize a kernel according to the unit sphere
do.stratified.cv.data
generates data for the stratified cross-validation. In particular subdivides the indices that refer to the rows of the data matrix in different folds (separated for positive and negative examples)
do.cv.data
generates data for the vanilla not stratified cross-validation.
labelsfromscores
computes the labels of a single class from the corresponding scores
Multiple.labels.from.scores
computes the labels of multiple classes from the corresponding scores
selection.test
is a non parametric test to select the most significant unlabeled examples
Value
compute.acc
returns the accuracy
compute.F
returns the F-score
norm1
returns the L1-norm value
Unit.sphere.norm
returns the kernel normalized according to the unit sphere
do.stratified.cv.data
returns a list with 2 two components:
fold.non.positives |
a list with k components. Each component is a vector with the indices of the non positive elements of the fold |
fold.positives |
a list with k components. Each component is a vector with the indices of the positive elements of the fold |
Indices refer to row numbers of the data matrix
do.cv.data
returns a list with 2 two components:
fold.non.positives |
a list with k components. Each component is a vector with the indices of the non positive elements of the fold |
fold.positives |
a list with k components. Each component is a vector with the indices of the positive elements of the fold |
Indices refer to row numbers of the data matrix
labelsfromscores
returns a numeric vector res with 0 or 1 values. The label res[i]=1 if scores[i]>thresh, otherwise res[i]=0
Multiple.labels.from.scores
returns a binary matrix with the labels of the predictions. Rows represent examples, columns classes. Element L[i,j] is the label of example i w.r.t. class j. L[i,j]=1 if i belongs to j, 0 otherwise.
selection.test
returns a list with 5 components:
selected |
a named vector with the components of av.scores selected by the test |
selected.labeled |
a named vector with the labeled components of av.scores selected by the test |
selected.unlabeled |
a named vector with the unlabeled components of av.scores selected by the test |
thresh |
the score threshold selected by the test |
alpha |
significance level (the same value of the input) |
Examples
# L1-norm of a vector
norm1(rnorm(10));
# generation of 5 stratified folds;
do.stratified.cv.data(1:100, 1:10, k = 5, seed = NULL);
# generation of labels form scores.
labelsfromscores(runif(20), thresh=0.3);