APML: Develop models with grid search

In APML: An Approach for Machine-Learning Modelling

Develop models with grid search

Description

Develop models with the optional parameters identified through the grid search and return model performance metrics. In order to run properly, the response column must be the first column and of a type of either numeric for "gaussian" or factor for "bernoulli" or "multinomial".

Usage

APML(model,AUC_stopping, xcol, traindata,testdata, hyper, 
distribution,imbalance, sort_by, extra_data,stopping_metric)

Arguments

model	The model to be used. Currently, only allow "gbm" (default) for the gradient boosted tree, and "rf" for the random forest.
AUC_stopping	Logical. If TRUE, the combinations of the hyperparameters will be randomly searched with AUC metric-based early stopping. Default:FALSE.
xcol	A vector containing the names or indices of the predictors to be used.
traindata	The training dataset.
testdata	The testing dataset.
hyper	List of hyper parameters (i.e., list(ntrees=c(1,2), max_depth=c(5,7)))
distribution	Distribution of the outcome: "bernoulli" (default), "bernoulli", "quasibinomial", "multinomial", "gaussian", "poisson", "gamma", "tweedie", "laplace", "quantile", "huber" or "custom".
imbalance	Logical. If true, balancing the case numbers in the training data via over/under-sampling when developing the model. Default:FALSE
sort_by	Select the best model in the grid space by sorting with a metric. Choices are "logloss", "residual_deviance", "mse", "auc", "accuracy", "precision", "recall", "f1", etc
extra_data	Extra dataset for evaluating model performance.
stopping_metric	Metric to use for early stopping (AUTO: logloss for classification, deviance for regression and anonomaly_score for Isolation Forest). Must be one of: "AUTO", "deviance", "logloss", "MSE", "RMSE", "MAE", "RMSLE", "AUC", "AUCPR", "lift_top_group", "misclassification", "mean_per_class_error", "custom", "custom_increasing". Defaults to AUTO.

Details

This function uses the grid search technique to tune models' parameters and return the optimal model.

Value

bestmodel	Best H2o model via grid search
train_metrics	Model performance metrics based on the training data.
test_metrics	Model performance metrics based on the testing data.
summary	Summary of model performance.
extra_metrics	Model performance metrics based on extra data. Only available when "model_metric" is used.

Note

This function is based on h2o package. In order to run this function, we need to run h2o.init() before using this function. The response variable should be the first column.

References

LeDell E, Gill N, Aiello S, Fu A, Candel A, Click C, et al. 2019. h2o: R Interface for “H2O.” Zhang W, Du Z, Zhang D, Yu S, Hao Y. 2016a. Boosted regression tree model-based assessment of the impacts of meteorological drivers of hand, foot and mouth disease in Guangdong, China. Sci Total Environ 553; doi:10.1016/j.scitotenv.2016.02.023.

Examples

library(h2o)
data(iris)
attach(iris)
h2o.init()
hyper <- list(ntrees=c(2,3,5))
iris <- iris[1:100,c(5,1:4)]
idx <- sample(100,50)
traindata <- iris[idx,]
testdata <- iris[-idx,]
xcol <- names(iris)[2:5]
results <- APML(xcol=xcol,hyper=hyper,
                traindata=traindata,testdata=testdata,
                sort_by ='auc',distribution = 'bernoulli')
h2o.shutdown(prompt=FALSE)
Sys.sleep(2)

APML documentation built on May 12, 2022, 9:06 a.m.