Description Usage Arguments Value Examples
The objective is - given the number of features, select the most informative ones and evaluate the predictive logistic regression model. The feature and model selection performed independently for each round of LOOCV. Feature selection performed using LASSO approach.
1 2 3 4 5 6 7 8 9 | lr_modeling(
msnset,
features,
response,
pred.cls,
K = NULL,
sel.feat = T,
par.backend = c("mc", "foreach", "none")
)
|
msnset |
MSnSet object. Note - can it be generalized to eset? |
features |
character vector features to select from for building prediction model. The features can be either in featureNames(msnset) or in pData(msnset). |
response |
factor to classify along. Must be only 2 levels. |
pred.cls |
character, class to predict |
K |
specifies the cross-validation type. Default NULL means LOOCV. Another typical value is 10. |
sel.feat |
logical to select features using LASSO or use the entire set? |
par.backend |
type of backend to support parallelizattion. 'mc' uses mclapply from parallel, 'foreach' is based on 'foreach', 'none' - just a single thread. |
list
prob
is the probabilities (response) from LOOCV that the sample is "case". That is how well model trained on other samples, predicts this particular one.
features
list of selected features for each iteration of LOOCV
top
top features over all iterations
auc
AUC
pred
prediction perfomance obtained by
ROCR::prediction
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 | data(srm_msnset)
head(varLabels(msnset))
head(msnset$subject.type)
# reduce to two classes
msnset <- msnset[,msnset$subject.type != "control.1"]
msnset$subject.type <- as.factor(msnset$subject.type)
# Note, par.backend="none" is for the example only.
out <- lr_modeling(msnset,
features=featureNames(msnset),
response="subject.type",
pred.cls="case", par.backend="none")
plotAUC(out)
# top features consistently recurring in the models during LOOCV
print(out$top)
# the AUC
print(out$auc)
# probabilities of classifying the sample right, if the feature selection
# and model training was performed on other samples
plot(sort(out$prob))
abline(h=0.5, col='red')
|
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.