trans_classifier: Create 'trans_classifier' object for machine-learning-based...
In microeco: Microbial Community Ecology Data Analysis
trans_classifier R Documentation
Create trans_classifier object for machine-learning-based model prediction.
Description
This class is a wrapper for methods of machine-learning-based classification or regression models, including data pre-processing, feature selection,
data split, model training, prediction, confusionMatrix and ROC (Receiver Operator Characteristic) or PR (Precision-Recall) curve.
Author(s): Felipe Mansoldo and Chi Liu
Methods
Public methods
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
Method new()
Create a trans_classifier object.
Usage
trans_classifier$new(
dataset,
x.predictors = "Genus",
y.response = NULL,
n.cores = 1
)
Arguments
datasetan object of microtable class.
x.predictorsdefault "Genus"; character vector or data.frame; a character vector represents selecting the corresponding data from microtable$taxa_abund;
data.frame denotes other customized input. See the following available options:
- 'Genus'
use Genus level table in microtable$taxa_abund, or other specific taxonomic rank, e.g., 'Phylum'.
At most one input level (e.g., ASV) is allowed to be absent in the names of taxa_abund list;
in that case the function will use otu_table to calculate relative abundance of features.
- 'all'
use all the levels stored in microtable$taxa_abund. Only available when it is the sole input.
- multiple levels
use several levels at the same time, e.g., c("Genus", "Species");
the corresponding tables in microtable$taxa_abund are combined by rows as the predictors.
- other input
must be a data.frame object. It should have the same format with the tables in microtable$taxa_abund, i.e. rows are features;
columns are samples with same names in sample_table.
y.responsedefault NULL; the response variable in sample_table of input microtable object.
n.coresdefault 1; the number of CPU threads used in the model training and the feature selection.
The parallel backend of the foreach package is registered only temporarily during the computation and the
original backend is restored afterwards, so that the global parallel setting is not permanently changed.
Returns
data_feature and data_response stored in the object.
Examples
\donttest{
data(dataset)
t1 <- trans_classifier$new(
dataset = dataset,
x.predictors = "Genus",
y.response = "Group")
# use multiple taxonomic levels as the predictors
t1 <- trans_classifier$new(
dataset = dataset,
x.predictors = c("Genus", "Species"),
y.response = "Group")
}
Method cal_split()
Split data for training and testing.
Usage
trans_classifier$cal_split(prop.train = 3/4)
Arguments
prop.traindefault 3/4; the ratio of the data used for the training.
Returns
data_train and data_test in the object.
Examples
\dontrun{
t1$cal_split(prop.train = 3/4)
}
Method cal_preProcess()
Pre-process (centering, scaling etc.) of features based on the caret::preProcess function.
See https://topepo.github.io/caret/pre-processing.html for more details.
Some methods of preProcess, such as 'nzv', 'zv', 'pca' and 'ica', change the number of features.
In that case the data_train and data_test are rebuilt with the new features and the
corresponding information is printed, so that the feature names always match the data.
The existing data_preProcess is reused when this function is called more than once; please assign
NULL to object$data_preProcess if a new preprocess model is needed.
Usage
trans_classifier$cal_preProcess(...)
Arguments
...parameters pass to preProcess function of caret package.
Returns
data_preProcess, data_train and data_test in the object.
data_preProcess is the return data generated by the preProcess function of caret package based on the training data.
data_train and data_test are preprocessed training and testing data based on the data_preProcess.
Examples
\dontrun{
# "nzv" removes near zero variance predictors
t1$cal_preProcess(method = c("center", "scale", "nzv"))
}
Method cal_feature_sel()
Perform feature selection.
See https://topepo.github.io/caret/feature-selection-overview.html for more details.
An error is raised when no feature is selected, as the following model training can not be performed
in that case. Please consider increasing boruta.repetitions, decreasing boruta.pValue or
using another x.predictors.
Usage
trans_classifier$cal_feature_sel(
boruta.maxRuns = 300,
boruta.pValue = 0.01,
boruta.repetitions = 4,
boruta.min.repetitions = 2,
seed = NULL,
...
)
Arguments
boruta.maxRunsdefault 300; maximal number of importance source runs; passed to the maxRuns parameter in Boruta function of Boruta package.
Note that the Boruta function requires a value greater than 10.
boruta.pValuedefault 0.01; p value passed to the pValue parameter in Boruta function of Boruta package.
boruta.repetitionsdefault 4; repetition runs for the feature selection.
boruta.min.repetitionsdefault 2; a feature is selected only when it is confirmed as important in at least
this number of the boruta.repetitions independent runs. It must not be greater than boruta.repetitions.
seeddefault NULL; the random seed used to make the repeated Boruta runs reproducible. When it is NULL,
the random seeds are generated automatically and the result may change between the runs.
...parameters pass to Boruta function of Boruta package.
Returns
optimized data_train and data_test in the object.
Examples
\dontrun{
t1$cal_feature_sel(boruta.maxRuns = 300, boruta.pValue = 0.01)
}
Method set_trainControl()
Control parameters for the following training. Please see trainControl function of caret package for details.
Usage
trans_classifier$set_trainControl(
method = "repeatedcv",
classProbs = TRUE,
savePredictions = "final",
...
)
Arguments
methoddefault 'repeatedcv'; 'repeatedcv': Repeated k-Fold cross validation;
see method parameter in trainControl function of caret package for available options.
classProbsdefault TRUE; should class probabilities be computed for classification models?;
see classProbs parameter in caret::trainControl function.
savePredictionsdefault 'final'; see savePredictions parameter in caret::trainControl function.
'final' means that only the predictions of the final model with the best tuning parameters are saved, which is
required by the cal_ROC function when input = "train". TRUE (or "all") saves the predictions of
all the tuning parameter combinations and is not recommended, as the repeated predictions of the same sample
are not independent observations and would distort the ROC analysis.
...parameters pass to trainControl function of caret package.
Returns
trainControl in the object.
Examples
\dontrun{
t1$set_trainControl(method = 'repeatedcv')
}
Method cal_train()
Run the model training. Please see https://topepo.github.io/caret/available-models.html for available models.
The mtry and ntree optimization is performed for the "rf" method in both the classification and the
regression. The mean Accuracy of the resamples is used to select the best ntree for the classification and
the mean RMSE for the regression. The selected mtry and ntree are stored in object$train_params.
Usage
trans_classifier$cal_train(method = "rf", max.mtry = 2, ntree = 500, ...)
Arguments
methoddefault "rf"; "rf": random forest; see method in train function of caret package for other options.
For method = "rf", the tuneGrid is set: expand.grid(mtry = seq(from = 1, to = max.mtry))
max.mtrydefault 2; for method = "rf"; maximum mtry used in the tuneGrid to do hyperparameter tuning to optimize the model.
ntreedefault 500; for method = "rf"; Number of trees to grow.
The default 500 is same with the ntree parameter in randomForest function in randomForest package.
When it is a vector with more than one element, the function will try to optimize the model to select a best one, such as c(100, 500, 1000).
...parameters pass to caret::train function.
Returns
res_train in the object.
Examples
\dontrun{
# random forest
t1$cal_train(method = "rf")
# Support Vector Machines with Radial Basis Function Kernel
t1$cal_train(method = "svmRadial", tuneLength = 15)
}
Method cal_feature_imp()
Get feature importance from the training model.
Usage
trans_classifier$cal_feature_imp(rf_feature_sig = FALSE, ...)
Arguments
rf_feature_sigdefault FALSE; whether calculate feature significance in 'rf' model using rfPermute package;
only available for method = "rf" in cal_train function.
...parameters pass to varImp function of caret package.
If rf_feature_sig is TRUE and train_method is "rf", the parameters will be passed to rfPermute function of rfPermute package.
Returns
res_feature_imp in the object. One row for each predictor variable. The column(s) are different importance measures.
When rf_feature_sig = FALSE, the importance is calculated with the varImp function of caret package and
the column is usually 'Overall' for both the classification and the regression. When rf_feature_sig = TRUE,
the importance values generated by the rfPermute package are used, such as 'MeanDecreaseAccuracy' and
'MeanDecreaseGini' for the classification, and the corresponding p values are stored in the columns with the
'.pval' suffix.
Examples
\dontrun{
t1$cal_feature_imp()
}
Method plot_feature_imp()
Bar plot for feature importance.
Usage
trans_classifier$plot_feature_imp(
rf_sig_show = NULL,
show_sig_group = FALSE,
...
)
Arguments
rf_sig_showdefault NULL; "MeanDecreaseAccuracy" (Default) or "MeanDecreaseGini" for random forest classification;
"%IncMSE" (Default) or "IncNodePurity" for random forest regression;
Only available when rf_feature_sig = TRUE in function cal_feature_imp,
which generate "MeanDecreaseGini" (and "MeanDecreaseAccuracy") or "%IncMSE" (and "IncNodePurity") in the column names of res_feature_imp;
Function can also generate "Significance" according to the p value.
show_sig_groupdefault FALSE; whether show the features with different significant groups;
Only available when "Significance" is found in the data.
...parameters pass to plot_diff_bar function of trans_diff package.
Returns
ggplot2 object.
Examples
\dontrun{
t1$plot_feature_imp(use_number = 1:20, coord_flip = FALSE)
}
Method cal_predict()
Run the prediction.
Usage
trans_classifier$cal_predict(positive_class = NULL)
Arguments
positive_classdefault NULL; see positive parameter in confusionMatrix function of caret package;
If positive_class is NULL, use the first group in data as the positive class automatically.
Returns
res_predict and the performance data stored in the object.
The res_predict is the predicted result for data_test.
For the classification, res_confusion_fit and res_confusion_stats are stored in the object.
For the regression, res_regression_stats with RMSE, Rsquared and MAE is stored in the object.
Several evaluation metrics in res_confusion_fit are defined as follows:
Accuracy = \frac{TP + TN}{TP + TN + FP + FN}
Sensitivity = Recall = TPR = \frac{TP}{TP + FN}
Specificity = TNR = 1 - FPR = \frac{TN}{TN + FP}
Precision = \frac{TP}{TP + FP}
Prevalence = \frac{TP + FN}{TP + TN + FP + FN}
F1-Score = \frac{2 * Precision * Recall}{Precision + Recall}
Kappa = \frac{Accuracy - Pe}{1 - Pe}
where TP is true positive; TN is ture negative; FP is false positive; and FN is false negative;
FPR is False Positive Rate; TPR is True Positive Rate; TNR is True Negative Rate;
Pe is the hypothetical probability of chance agreement on the classes for reference and prediction in the confusion matrix.
Accuracy represents the ratio of correct predictions.
Precision identifies how the model accurately predicted the positive classes.
Recall (sensitivity) measures the ratio of actual positives that are correctly identified by the model.
F1-score is the weighted average score of recall and precision. The value at 1 is the best performance and at 0 is the worst.
Prevalence represents how often positive events occurred.
Kappa identifies how well the model is predicting.
Examples
\dontrun{
t1$cal_predict()
}
Method plot_confusionMatrix()
Plot the cross-tabulation of observed and predicted classes with associated statistics based on the results of function cal_predict.
Usage
trans_classifier$plot_confusionMatrix(
plot_confusion = TRUE,
plot_statistics = TRUE,
as_ggplot = FALSE
)
Arguments
plot_confusiondefault TRUE; whether plot the confusion matrix.
plot_statisticsdefault TRUE; whether plot the statistics.
as_ggplotdefault FALSE; whether return a ggplot object instead of the gtable object
assembled by the gridExtra package. As the statistics table is plotted with the tableGrob function,
only the confusion matrix is returned when as_ggplot = TRUE; the statistics table is available in
object$res_confusion_stats.
Returns
gtable object assembled by the gridExtra package when as_ggplot = FALSE (default),
or ggplot object when as_ggplot = TRUE.
Examples
\dontrun{
t1$plot_confusionMatrix()
}
Method cal_ROC()
Get ROC (Receiver Operator Characteristic) curve data and the performance data.
When input = "train", the predictions are obtained from object$res_train$pred, which requires
savePredictions in set_trainControl to be 'final' (default). When the predictions of several
tuning parameter combinations are saved (e.g. savePredictions = TRUE), only the first prediction of
each sample is used, as the repeated predictions of a sample are not independent observations.
Usage
trans_classifier$cal_ROC(input = "pred")
Arguments
inputdefault "pred"; 'pred' or 'train'; 'pred' represents using prediction results;
'train' represents using training results.
Returns
a list res_ROC stored in the object. It has two tables: res_roc and res_pr. AUC: Area Under the ROC Curve.
For the definition of metrics, please refer to the return part of function cal_predict.
Examples
\dontrun{
t1$cal_ROC()
}
Method plot_ROC()
Plot ROC curve.
Usage
trans_classifier$plot_ROC(
plot_type = c("ROC", "PR")[1],
plot_group = "all",
color_values = RColorBrewer::brewer.pal(8, "Dark2"),
add_AUC = TRUE,
plot_method = FALSE,
...
)
Arguments
plot_typedefault c("ROC", "PR")[1]; 'ROC' represents ROC (Receiver Operator Characteristic) curve;
'PR' represents PR (Precision-Recall) curve.
plot_groupdefault "all"; 'all' represents all the classes in the model;
'add' represents all adding micro-average and macro-average results, see
https://scikit-learn.org/stable/auto_examples/model_selection/plot_roc.html;
other options should be one or more class names, same with the names in Group column of res_ROC$res_roc from cal_ROC function.
color_valuesdefault RColorBrewer::brewer.pal(8, "Dark2"); colors used in the plot.
add_AUCdefault TRUE; whether add AUC in the legend.
plot_methoddefault FALSE; If TRUE, show the method in the legend though only one method is found.
...parameters pass to geom_path function of ggplot2 package.
Returns
ggplot2 object.
Examples
\dontrun{
t1$plot_ROC(size = 1, alpha = 0.7)
}
Method cal_caretList()
Use caretList function of caretEnsemble package to run multiple models. For the available models, please run names(getModelInfo()).
Usage
trans_classifier$cal_caretList(...)
Arguments
...parameters pass to caretList function of caretEnsemble package.
Returns
res_caretList_models in the object.
Examples
\dontrun{
t1$cal_caretList(methodList = c('rf', 'svmRadial'))
}
Method cal_caretList_resamples()
Use resamples function of caret package to collect the metric values based on the res_caretList_models data.
Usage
trans_classifier$cal_caretList_resamples(...)
Arguments
...parameters pass to resamples function of caret package.
Returns
res_caretList_resamples list and res_caretList_resamples_reshaped table in the object.
Examples
\dontrun{
t1$cal_caretList_resamples()
}
Method plot_caretList_resamples()
Visualize the metric values based on the res_caretList_resamples_reshaped data.
Usage
trans_classifier$plot_caretList_resamples(
color_values = RColorBrewer::brewer.pal(8, "Dark2"),
...
)
Arguments
color_valuesdefault RColorBrewer::brewer.pal(8, "Dark2"); colors palette for the box.
...parameters pass to geom_boxplot function of ggplot2 package.
Returns
ggplot object.
Examples
\dontrun{
t1$plot_caretList_resamples()
}
Method print()
Print the trans_classifier object.
Usage
trans_classifier$print()
Method clone()
The objects of this class are cloneable with this method.
Usage
trans_classifier$clone(deep = FALSE)
Arguments
deepWhether to make a deep clone.
Examples
## ------------------------------------------------
## Method `trans_classifier$new`
## ------------------------------------------------
data(dataset)
t1 <- trans_classifier$new(
dataset = dataset,
x.predictors = "Genus",
y.response = "Group")
# use multiple taxonomic levels as the predictors
t1 <- trans_classifier$new(
dataset = dataset,
x.predictors = c("Genus", "Species"),
y.response = "Group")
## ------------------------------------------------
## Method `trans_classifier$cal_split`
## ------------------------------------------------
## Not run:
t1$cal_split(prop.train = 3/4)
## End(Not run)
## ------------------------------------------------
## Method `trans_classifier$cal_preProcess`
## ------------------------------------------------
## Not run:
# "nzv" removes near zero variance predictors
t1$cal_preProcess(method = c("center", "scale", "nzv"))
## End(Not run)
## ------------------------------------------------
## Method `trans_classifier$cal_feature_sel`
## ------------------------------------------------
## Not run:
t1$cal_feature_sel(boruta.maxRuns = 300, boruta.pValue = 0.01)
## End(Not run)
## ------------------------------------------------
## Method `trans_classifier$set_trainControl`
## ------------------------------------------------
## Not run:
t1$set_trainControl(method = 'repeatedcv')
## End(Not run)
## ------------------------------------------------
## Method `trans_classifier$cal_train`
## ------------------------------------------------
## Not run:
# random forest
t1$cal_train(method = "rf")
# Support Vector Machines with Radial Basis Function Kernel
t1$cal_train(method = "svmRadial", tuneLength = 15)
## End(Not run)
## ------------------------------------------------
## Method `trans_classifier$cal_feature_imp`
## ------------------------------------------------
## Not run:
t1$cal_feature_imp()
## End(Not run)
## ------------------------------------------------
## Method `trans_classifier$plot_feature_imp`
## ------------------------------------------------
## Not run:
t1$plot_feature_imp(use_number = 1:20, coord_flip = FALSE)
## End(Not run)
## ------------------------------------------------
## Method `trans_classifier$cal_predict`
## ------------------------------------------------
## Not run:
t1$cal_predict()
## End(Not run)
## ------------------------------------------------
## Method `trans_classifier$plot_confusionMatrix`
## ------------------------------------------------
## Not run:
t1$plot_confusionMatrix()
## End(Not run)
## ------------------------------------------------
## Method `trans_classifier$cal_ROC`
## ------------------------------------------------
## Not run:
t1$cal_ROC()
## End(Not run)
## ------------------------------------------------
## Method `trans_classifier$plot_ROC`
## ------------------------------------------------
## Not run:
t1$plot_ROC(size = 1, alpha = 0.7)
## End(Not run)
## ------------------------------------------------
## Method `trans_classifier$cal_caretList`
## ------------------------------------------------
## Not run:
t1$cal_caretList(methodList = c('rf', 'svmRadial'))
## End(Not run)
## ------------------------------------------------
## Method `trans_classifier$cal_caretList_resamples`
## ------------------------------------------------
## Not run:
t1$cal_caretList_resamples()
## End(Not run)
## ------------------------------------------------
## Method `trans_classifier$plot_caretList_resamples`
## ------------------------------------------------
## Not run:
t1$plot_caretList_resamples()
## End(Not run)
microeco documentation built on Oct. 20, 2026, 9:06 a.m.
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
| trans_classifier | R Documentation |
Create trans_classifier object for machine-learning-based model prediction.
Description
This class is a wrapper for methods of machine-learning-based classification or regression models, including data pre-processing, feature selection, data split, model training, prediction, confusionMatrix and ROC (Receiver Operator Characteristic) or PR (Precision-Recall) curve.
Author(s): Felipe Mansoldo and Chi Liu
Methods
Public methods
Method new()
Create a trans_classifier object.
Usage
trans_classifier$new( dataset, x.predictors = "Genus", y.response = NULL, n.cores = 1 )
Arguments
datasetan object of
microtableclass.x.predictorsdefault "Genus"; character vector or data.frame; a character vector represents selecting the corresponding data from
microtable$taxa_abund; data.frame denotes other customized input. See the following available options:- 'Genus'
use Genus level table in
microtable$taxa_abund, or other specific taxonomic rank, e.g., 'Phylum'. At most one input level (e.g., ASV) is allowed to be absent in the names of taxa_abund list; in that case the function will useotu_tableto calculate relative abundance of features.- 'all'
use all the levels stored in
microtable$taxa_abund. Only available when it is the sole input.- multiple levels
use several levels at the same time, e.g.,
c("Genus", "Species"); the corresponding tables inmicrotable$taxa_abundare combined by rows as the predictors.- other input
must be a data.frame object. It should have the same format with the tables in microtable$taxa_abund, i.e. rows are features; columns are samples with same names in sample_table.
y.responsedefault NULL; the response variable in
sample_tableof inputmicrotableobject.n.coresdefault 1; the number of CPU threads used in the model training and the feature selection. The parallel backend of the foreach package is registered only temporarily during the computation and the original backend is restored afterwards, so that the global parallel setting is not permanently changed.
Returns
data_feature and data_response stored in the object.
Examples
\donttest{
data(dataset)
t1 <- trans_classifier$new(
dataset = dataset,
x.predictors = "Genus",
y.response = "Group")
# use multiple taxonomic levels as the predictors
t1 <- trans_classifier$new(
dataset = dataset,
x.predictors = c("Genus", "Species"),
y.response = "Group")
}
Method cal_split()
Split data for training and testing.
Usage
trans_classifier$cal_split(prop.train = 3/4)
Arguments
prop.traindefault 3/4; the ratio of the data used for the training.
Returns
data_train and data_test in the object.
Examples
\dontrun{
t1$cal_split(prop.train = 3/4)
}
Method cal_preProcess()
Pre-process (centering, scaling etc.) of features based on the caret::preProcess function.
See https://topepo.github.io/caret/pre-processing.html for more details.
Some methods of preProcess, such as 'nzv', 'zv', 'pca' and 'ica', change the number of features.
In that case the data_train and data_test are rebuilt with the new features and the
corresponding information is printed, so that the feature names always match the data.
The existing data_preProcess is reused when this function is called more than once; please assign
NULL to object$data_preProcess if a new preprocess model is needed.
Usage
trans_classifier$cal_preProcess(...)
Arguments
...parameters pass to
preProcessfunction of caret package.
Returns
data_preProcess, data_train and data_test in the object.
data_preProcess is the return data generated by the preProcess function of caret package based on the training data.
data_train and data_test are preprocessed training and testing data based on the data_preProcess.
Examples
\dontrun{
# "nzv" removes near zero variance predictors
t1$cal_preProcess(method = c("center", "scale", "nzv"))
}
Method cal_feature_sel()
Perform feature selection.
See https://topepo.github.io/caret/feature-selection-overview.html for more details.
An error is raised when no feature is selected, as the following model training can not be performed
in that case. Please consider increasing boruta.repetitions, decreasing boruta.pValue or
using another x.predictors.
Usage
trans_classifier$cal_feature_sel( boruta.maxRuns = 300, boruta.pValue = 0.01, boruta.repetitions = 4, boruta.min.repetitions = 2, seed = NULL, ... )
Arguments
boruta.maxRunsdefault 300; maximal number of importance source runs; passed to the
maxRunsparameter inBorutafunction of Boruta package. Note that the Boruta function requires a value greater than 10.boruta.pValuedefault 0.01; p value passed to the pValue parameter in
Borutafunction of Boruta package.boruta.repetitionsdefault 4; repetition runs for the feature selection.
boruta.min.repetitionsdefault 2; a feature is selected only when it is confirmed as important in at least this number of the
boruta.repetitionsindependent runs. It must not be greater thanboruta.repetitions.seeddefault NULL; the random seed used to make the repeated Boruta runs reproducible. When it is NULL, the random seeds are generated automatically and the result may change between the runs.
...parameters pass to
Borutafunction of Boruta package.
Returns
optimized data_train and data_test in the object.
Examples
\dontrun{
t1$cal_feature_sel(boruta.maxRuns = 300, boruta.pValue = 0.01)
}
Method set_trainControl()
Control parameters for the following training. Please see trainControl function of caret package for details.
Usage
trans_classifier$set_trainControl( method = "repeatedcv", classProbs = TRUE, savePredictions = "final", ... )
Arguments
methoddefault 'repeatedcv'; 'repeatedcv': Repeated k-Fold cross validation; see method parameter in
trainControlfunction ofcaretpackage for available options.classProbsdefault TRUE; should class probabilities be computed for classification models?; see classProbs parameter in
caret::trainControlfunction.savePredictionsdefault 'final'; see
savePredictionsparameter incaret::trainControlfunction. 'final' means that only the predictions of the final model with the best tuning parameters are saved, which is required by thecal_ROCfunction wheninput = "train". TRUE (or "all") saves the predictions of all the tuning parameter combinations and is not recommended, as the repeated predictions of the same sample are not independent observations and would distort the ROC analysis....parameters pass to
trainControlfunction of caret package.
Returns
trainControl in the object.
Examples
\dontrun{
t1$set_trainControl(method = 'repeatedcv')
}
Method cal_train()
Run the model training. Please see https://topepo.github.io/caret/available-models.html for available models.
The mtry and ntree optimization is performed for the "rf" method in both the classification and the
regression. The mean Accuracy of the resamples is used to select the best ntree for the classification and
the mean RMSE for the regression. The selected mtry and ntree are stored in object$train_params.
Usage
trans_classifier$cal_train(method = "rf", max.mtry = 2, ntree = 500, ...)
Arguments
methoddefault "rf"; "rf": random forest; see method in
trainfunction of caret package for other options. For method = "rf", thetuneGridis set:expand.grid(mtry = seq(from = 1, to = max.mtry))max.mtrydefault 2; for method = "rf"; maximum mtry used in the
tuneGridto do hyperparameter tuning to optimize the model.ntreedefault 500; for method = "rf"; Number of trees to grow. The default 500 is same with the
ntreeparameter inrandomForestfunction in randomForest package. When it is a vector with more than one element, the function will try to optimize the model to select a best one, such asc(100, 500, 1000)....parameters pass to
caret::trainfunction.
Returns
res_train in the object.
Examples
\dontrun{
# random forest
t1$cal_train(method = "rf")
# Support Vector Machines with Radial Basis Function Kernel
t1$cal_train(method = "svmRadial", tuneLength = 15)
}
Method cal_feature_imp()
Get feature importance from the training model.
Usage
trans_classifier$cal_feature_imp(rf_feature_sig = FALSE, ...)
Arguments
rf_feature_sigdefault FALSE; whether calculate feature significance in 'rf' model using
rfPermutepackage; only available formethod = "rf"incal_trainfunction....parameters pass to
varImpfunction of caret package. Ifrf_feature_sigis TRUE andtrain_methodis "rf", the parameters will be passed torfPermutefunction of rfPermute package.
Returns
res_feature_imp in the object. One row for each predictor variable. The column(s) are different importance measures.
When rf_feature_sig = FALSE, the importance is calculated with the varImp function of caret package and
the column is usually 'Overall' for both the classification and the regression. When rf_feature_sig = TRUE,
the importance values generated by the rfPermute package are used, such as 'MeanDecreaseAccuracy' and
'MeanDecreaseGini' for the classification, and the corresponding p values are stored in the columns with the
'.pval' suffix.
Examples
\dontrun{
t1$cal_feature_imp()
}
Method plot_feature_imp()
Bar plot for feature importance.
Usage
trans_classifier$plot_feature_imp( rf_sig_show = NULL, show_sig_group = FALSE, ... )
Arguments
rf_sig_showdefault NULL; "MeanDecreaseAccuracy" (Default) or "MeanDecreaseGini" for random forest classification; "%IncMSE" (Default) or "IncNodePurity" for random forest regression; Only available when
rf_feature_sig = TRUEin functioncal_feature_imp, which generate "MeanDecreaseGini" (and "MeanDecreaseAccuracy") or "%IncMSE" (and "IncNodePurity") in the column names ofres_feature_imp; Function can also generate "Significance" according to the p value.show_sig_groupdefault FALSE; whether show the features with different significant groups; Only available when "Significance" is found in the data.
...parameters pass to
plot_diff_barfunction oftrans_diffpackage.
Returns
ggplot2 object.
Examples
\dontrun{
t1$plot_feature_imp(use_number = 1:20, coord_flip = FALSE)
}
Method cal_predict()
Run the prediction.
Usage
trans_classifier$cal_predict(positive_class = NULL)
Arguments
positive_classdefault NULL; see positive parameter in
confusionMatrixfunction of caret package; If positive_class is NULL, use the first group in data as the positive class automatically.
Returns
res_predict and the performance data stored in the object.
The res_predict is the predicted result for data_test.
For the classification, res_confusion_fit and res_confusion_stats are stored in the object.
For the regression, res_regression_stats with RMSE, Rsquared and MAE is stored in the object.
Several evaluation metrics in res_confusion_fit are defined as follows:
Accuracy = \frac{TP + TN}{TP + TN + FP + FN}
Sensitivity = Recall = TPR = \frac{TP}{TP + FN}
Specificity = TNR = 1 - FPR = \frac{TN}{TN + FP}
Precision = \frac{TP}{TP + FP}
Prevalence = \frac{TP + FN}{TP + TN + FP + FN}
F1-Score = \frac{2 * Precision * Recall}{Precision + Recall}
Kappa = \frac{Accuracy - Pe}{1 - Pe}
where TP is true positive; TN is ture negative; FP is false positive; and FN is false negative; FPR is False Positive Rate; TPR is True Positive Rate; TNR is True Negative Rate; Pe is the hypothetical probability of chance agreement on the classes for reference and prediction in the confusion matrix. Accuracy represents the ratio of correct predictions. Precision identifies how the model accurately predicted the positive classes. Recall (sensitivity) measures the ratio of actual positives that are correctly identified by the model. F1-score is the weighted average score of recall and precision. The value at 1 is the best performance and at 0 is the worst. Prevalence represents how often positive events occurred. Kappa identifies how well the model is predicting.
Examples
\dontrun{
t1$cal_predict()
}
Method plot_confusionMatrix()
Plot the cross-tabulation of observed and predicted classes with associated statistics based on the results of function cal_predict.
Usage
trans_classifier$plot_confusionMatrix( plot_confusion = TRUE, plot_statistics = TRUE, as_ggplot = FALSE )
Arguments
plot_confusiondefault TRUE; whether plot the confusion matrix.
plot_statisticsdefault TRUE; whether plot the statistics.
as_ggplotdefault FALSE; whether return a
ggplotobject instead of thegtableobject assembled by the gridExtra package. As the statistics table is plotted with thetableGrobfunction, only the confusion matrix is returned whenas_ggplot = TRUE; the statistics table is available inobject$res_confusion_stats.
Returns
gtable object assembled by the gridExtra package when as_ggplot = FALSE (default),
or ggplot object when as_ggplot = TRUE.
Examples
\dontrun{
t1$plot_confusionMatrix()
}
Method cal_ROC()
Get ROC (Receiver Operator Characteristic) curve data and the performance data.
When input = "train", the predictions are obtained from object$res_train$pred, which requires
savePredictions in set_trainControl to be 'final' (default). When the predictions of several
tuning parameter combinations are saved (e.g. savePredictions = TRUE), only the first prediction of
each sample is used, as the repeated predictions of a sample are not independent observations.
Usage
trans_classifier$cal_ROC(input = "pred")
Arguments
inputdefault "pred"; 'pred' or 'train'; 'pred' represents using prediction results; 'train' represents using training results.
Returns
a list res_ROC stored in the object. It has two tables: res_roc and res_pr. AUC: Area Under the ROC Curve.
For the definition of metrics, please refer to the return part of function cal_predict.
Examples
\dontrun{
t1$cal_ROC()
}
Method plot_ROC()
Plot ROC curve.
Usage
trans_classifier$plot_ROC(
plot_type = c("ROC", "PR")[1],
plot_group = "all",
color_values = RColorBrewer::brewer.pal(8, "Dark2"),
add_AUC = TRUE,
plot_method = FALSE,
...
)Arguments
plot_typedefault c("ROC", "PR")[1]; 'ROC' represents ROC (Receiver Operator Characteristic) curve; 'PR' represents PR (Precision-Recall) curve.
plot_groupdefault "all"; 'all' represents all the classes in the model; 'add' represents all adding micro-average and macro-average results, see https://scikit-learn.org/stable/auto_examples/model_selection/plot_roc.html; other options should be one or more class names, same with the names in Group column of res_ROC$res_roc from cal_ROC function.
color_valuesdefault RColorBrewer::brewer.pal(8, "Dark2"); colors used in the plot.
add_AUCdefault TRUE; whether add AUC in the legend.
plot_methoddefault FALSE; If TRUE, show the method in the legend though only one method is found.
...parameters pass to
geom_pathfunction of ggplot2 package.
Returns
ggplot2 object.
Examples
\dontrun{
t1$plot_ROC(size = 1, alpha = 0.7)
}
Method cal_caretList()
Use caretList function of caretEnsemble package to run multiple models. For the available models, please run names(getModelInfo()).
Usage
trans_classifier$cal_caretList(...)
Arguments
...parameters pass to
caretListfunction ofcaretEnsemblepackage.
Returns
res_caretList_models in the object.
Examples
\dontrun{
t1$cal_caretList(methodList = c('rf', 'svmRadial'))
}
Method cal_caretList_resamples()
Use resamples function of caret package to collect the metric values based on the res_caretList_models data.
Usage
trans_classifier$cal_caretList_resamples(...)
Arguments
...parameters pass to
resamplesfunction ofcaretpackage.
Returns
res_caretList_resamples list and res_caretList_resamples_reshaped table in the object.
Examples
\dontrun{
t1$cal_caretList_resamples()
}
Method plot_caretList_resamples()
Visualize the metric values based on the res_caretList_resamples_reshaped data.
Usage
trans_classifier$plot_caretList_resamples( color_values = RColorBrewer::brewer.pal(8, "Dark2"), ... )
Arguments
color_valuesdefault
RColorBrewer::brewer.pal(8, "Dark2"); colors palette for the box....parameters pass to
geom_boxplotfunction ofggplot2package.
Returns
ggplot object.
Examples
\dontrun{
t1$plot_caretList_resamples()
}
Method print()
Print the trans_classifier object.
Usage
trans_classifier$print()
Method clone()
The objects of this class are cloneable with this method.
Usage
trans_classifier$clone(deep = FALSE)
Arguments
deepWhether to make a deep clone.
Examples
## ------------------------------------------------
## Method `trans_classifier$new`
## ------------------------------------------------
data(dataset)
t1 <- trans_classifier$new(
dataset = dataset,
x.predictors = "Genus",
y.response = "Group")
# use multiple taxonomic levels as the predictors
t1 <- trans_classifier$new(
dataset = dataset,
x.predictors = c("Genus", "Species"),
y.response = "Group")
## ------------------------------------------------
## Method `trans_classifier$cal_split`
## ------------------------------------------------
## Not run:
t1$cal_split(prop.train = 3/4)
## End(Not run)
## ------------------------------------------------
## Method `trans_classifier$cal_preProcess`
## ------------------------------------------------
## Not run:
# "nzv" removes near zero variance predictors
t1$cal_preProcess(method = c("center", "scale", "nzv"))
## End(Not run)
## ------------------------------------------------
## Method `trans_classifier$cal_feature_sel`
## ------------------------------------------------
## Not run:
t1$cal_feature_sel(boruta.maxRuns = 300, boruta.pValue = 0.01)
## End(Not run)
## ------------------------------------------------
## Method `trans_classifier$set_trainControl`
## ------------------------------------------------
## Not run:
t1$set_trainControl(method = 'repeatedcv')
## End(Not run)
## ------------------------------------------------
## Method `trans_classifier$cal_train`
## ------------------------------------------------
## Not run:
# random forest
t1$cal_train(method = "rf")
# Support Vector Machines with Radial Basis Function Kernel
t1$cal_train(method = "svmRadial", tuneLength = 15)
## End(Not run)
## ------------------------------------------------
## Method `trans_classifier$cal_feature_imp`
## ------------------------------------------------
## Not run:
t1$cal_feature_imp()
## End(Not run)
## ------------------------------------------------
## Method `trans_classifier$plot_feature_imp`
## ------------------------------------------------
## Not run:
t1$plot_feature_imp(use_number = 1:20, coord_flip = FALSE)
## End(Not run)
## ------------------------------------------------
## Method `trans_classifier$cal_predict`
## ------------------------------------------------
## Not run:
t1$cal_predict()
## End(Not run)
## ------------------------------------------------
## Method `trans_classifier$plot_confusionMatrix`
## ------------------------------------------------
## Not run:
t1$plot_confusionMatrix()
## End(Not run)
## ------------------------------------------------
## Method `trans_classifier$cal_ROC`
## ------------------------------------------------
## Not run:
t1$cal_ROC()
## End(Not run)
## ------------------------------------------------
## Method `trans_classifier$plot_ROC`
## ------------------------------------------------
## Not run:
t1$plot_ROC(size = 1, alpha = 0.7)
## End(Not run)
## ------------------------------------------------
## Method `trans_classifier$cal_caretList`
## ------------------------------------------------
## Not run:
t1$cal_caretList(methodList = c('rf', 'svmRadial'))
## End(Not run)
## ------------------------------------------------
## Method `trans_classifier$cal_caretList_resamples`
## ------------------------------------------------
## Not run:
t1$cal_caretList_resamples()
## End(Not run)
## ------------------------------------------------
## Method `trans_classifier$plot_caretList_resamples`
## ------------------------------------------------
## Not run:
t1$plot_caretList_resamples()
## End(Not run)
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
Embedding an R snippet on your website
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.