Description Usage Arguments Value Examples
The objective is - given the number of features, select the most informative ones and evaluate the predictive random forest model. The feature and model selection performed independently for each round of LOOCV.
1 2 3 4 5 6 7 8 9 10 | rf_modeling(
msnset,
features,
response,
pred.cls,
K = NULL,
sel.feat = T,
sel.alg = c("varSelRF", "Boruta", "top"),
...
)
|
msnset |
MSnSet object |
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 |
class to predict |
K |
specifies the cross-validation type. Default NULL means LOOCV. Another typical value is 10. |
sel.feat |
logical defining if to select features or use the entire set? |
sel.alg |
character.
|
... |
Extra arguments. Currently passed only to Boruta algorithm. |
verbose |
0 - silent, 1 - count round, 2 - print selected features at each round |
list
prob
is the probabilities (response) from LOOCV that the sample is "case"
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 | 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)
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.