| Title: | Automated Stacked Ensemble Classifier for Severe Class Imbalance |
|---|---|
| Description: | An AutoML algorithm is developed to construct homogeneous or heterogeneous stacked ensemble models using specified base-learners. Various criteria are employed to identify optimal models, enhancing diversity among them and resulting in more robust stacked ensembles. The algorithm optimizes the model by incorporating an increasing number of top-performing models to create a diverse combination. Presently, only models from 'h2o.ai' are supported. |
| Authors: | E. F. Haghish [aut, cre, cph] |
| Maintainer: | E. F. Haghish <[email protected]> |
| License: | MIT + file LICENSE |
| Version: | 0.3 |
| Built: | 2025-01-26 02:55:08 UTC |
| Source: | https://github.com/haghish/autoensemble |
Automatically trains various algorithms to build base-learners and then automatically creates a stacked ensemble model
autoEnsemble( x, y, training_frame, validation_frame = NULL, nfolds = 10, balance_classes = TRUE, max_runtime_secs = NULL, max_runtime_secs_per_model = NULL, max_models = NULL, sort_metric = "AUCPR", include_algos = c("GLM", "DeepLearning", "DRF", "XGBoost", "GBM"), save_models = FALSE, directory = paste("autoEnsemble", format(Sys.time(), "%d-%m-%y-%H:%M")), zip = FALSE, verbosity = NULL, newdata = NULL, family = "binary", strategy = c("search"), model_selection_criteria = c("auc", "aucpr", "mcc", "f2"), min_improvement = 1e-05, max = NULL, top_rank = seq(0.01, 0.99, 0.01), stop_rounds = 3, reset_stop_rounds = TRUE, stop_metric = "auc", seed = -1, verbatim = FALSE, startH2O = FALSE, nthreads = NULL, max_mem_size = NULL, min_mem_size = NULL, ignore_config = FALSE, bind_to_localhost = FALSE, insecure = TRUE )autoEnsemble( x, y, training_frame, validation_frame = NULL, nfolds = 10, balance_classes = TRUE, max_runtime_secs = NULL, max_runtime_secs_per_model = NULL, max_models = NULL, sort_metric = "AUCPR", include_algos = c("GLM", "DeepLearning", "DRF", "XGBoost", "GBM"), save_models = FALSE, directory = paste("autoEnsemble", format(Sys.time(), "%d-%m-%y-%H:%M")), zip = FALSE, verbosity = NULL, newdata = NULL, family = "binary", strategy = c("search"), model_selection_criteria = c("auc", "aucpr", "mcc", "f2"), min_improvement = 1e-05, max = NULL, top_rank = seq(0.01, 0.99, 0.01), stop_rounds = 3, reset_stop_rounds = TRUE, stop_metric = "auc", seed = -1, verbatim = FALSE, startH2O = FALSE, nthreads = NULL, max_mem_size = NULL, min_mem_size = NULL, ignore_config = FALSE, bind_to_localhost = FALSE, insecure = TRUE )
training_frame |
h2o training frame (data.frame) for model training |
newdata |
h2o frame (data.frame). the data.frame must be already uploaded on h2o server (cloud). when specified, this dataset will be used for evaluating the models. if not specified, model performance on the training dataset will be reported. |
family |
model family. currently only |
strategy |
character. the current available strategies are |
model_selection_criteria |
character, specifying the performance metrics that
should be taken into consideration for model selection. the default are
|
min_improvement |
numeric. specifies the minimum improvement in model evaluation metric to qualify further optimization search. |
max |
integer. specifies maximum number of models for each criteria to be extracted. the
default value is the |
top_rank |
numeric vector. specifies percentage of the top models taht
should be selected. if the strategy is |
stop_rounds |
integer. number of stoping rounds, in case the model stops improving |
reset_stop_rounds |
logical. if TRUE, everytime the model improves the stopping rounds penalty is resets to 0. |
stop_metric |
character. model stopping metric. the default is |
seed |
random seed (recommended) |
verbatim |
logical. if TRUE, it reports additional information about the progress of the model training, particularly used for debugging. |
models |
H2O search grid or AutoML grid or a character vector of H2O model IDs.
the |
a list including the ensemble model and the top-rank models that were used in the model
E. F. Haghish
## Not run: # load the required libraries for building the base-learners and the ensemble models library(h2o) library(autoEnsemble) # initiate the h2o server h2o.init(ignore_config = TRUE, nthreads = 2, bind_to_localhost = FALSE, insecure = TRUE) # upload data to h2o cloud prostate_path <- system.file("extdata", "prostate.csv", package = "h2o") prostate <- h2o.importFile(path = prostate_path, header = TRUE) ### H2O provides 2 types of grid search for tuning the models, which are ### AutoML and Grid. Below, I tune 2 set of model grids and use them both ### for building the ensemble, just to set an example ... ####################################################### ### PREPARE AutoML Grid (takes a couple of minutes) ####################################################### # run AutoML to tune various models (GLM, GBM, XGBoost, DRF, DeepLearning) for 120 seconds y <- "CAPSULE" prostate[,y] <- as.factor(prostate[,y]) #convert to factor for classification aml <- h2o.automl(y = y, training_frame = prostate, max_runtime_secs = 120, include_algos=c("DRF","GLM", "XGBoost", "GBM", "DeepLearning"), # this setting ensures the models are comparable for building a meta learner seed = 2023, nfolds = 10, keep_cross_validation_predictions = TRUE) ####################################################### ### PREPARE H2O Grid (takes a couple of minutes) ####################################################### # make sure equal number of "nfolds" is specified for different grids grid <- h2o.grid(algorithm = "gbm", y = y, training_frame = prostate, hyper_params = list(ntrees = seq(1,50,1)), grid_id = "ensemble_grid", # this setting ensures the models are comparable for building a meta learner seed = 2023, fold_assignment = "Modulo", nfolds = 10, keep_cross_validation_predictions = TRUE) ####################################################### ### PREPARE ENSEMBLE MODEL ####################################################### ### get the models' IDs from the AutoML and grid searches. ### this is all that is needed before building the ensemble, ### i.e., to specify the model IDs that should be evaluated. ids <- c(h2o.get_ids(aml), h2o.get_ids(grid)) top <- ensemble(models = ids, training_frame = prostate, strategy = "top") search <- ensemble(models = ids, training_frame = prostate, strategy = "search") ####################################################### ### EVALUATE THE MODELS ####################################################### h2o.auc(aml@leader) # best model identified by h2o.automl h2o.auc(h2o.getModel(grid@model_ids[[1]])) # best model identified by grid search h2o.auc(top$model). # ensemble model with 'top' search strategy h2o.auc(search$model). # ensemble model with 'search' search strategy ## End(Not run)## Not run: # load the required libraries for building the base-learners and the ensemble models library(h2o) library(autoEnsemble) # initiate the h2o server h2o.init(ignore_config = TRUE, nthreads = 2, bind_to_localhost = FALSE, insecure = TRUE) # upload data to h2o cloud prostate_path <- system.file("extdata", "prostate.csv", package = "h2o") prostate <- h2o.importFile(path = prostate_path, header = TRUE) ### H2O provides 2 types of grid search for tuning the models, which are ### AutoML and Grid. Below, I tune 2 set of model grids and use them both ### for building the ensemble, just to set an example ... ####################################################### ### PREPARE AutoML Grid (takes a couple of minutes) ####################################################### # run AutoML to tune various models (GLM, GBM, XGBoost, DRF, DeepLearning) for 120 seconds y <- "CAPSULE" prostate[,y] <- as.factor(prostate[,y]) #convert to factor for classification aml <- h2o.automl(y = y, training_frame = prostate, max_runtime_secs = 120, include_algos=c("DRF","GLM", "XGBoost", "GBM", "DeepLearning"), # this setting ensures the models are comparable for building a meta learner seed = 2023, nfolds = 10, keep_cross_validation_predictions = TRUE) ####################################################### ### PREPARE H2O Grid (takes a couple of minutes) ####################################################### # make sure equal number of "nfolds" is specified for different grids grid <- h2o.grid(algorithm = "gbm", y = y, training_frame = prostate, hyper_params = list(ntrees = seq(1,50,1)), grid_id = "ensemble_grid", # this setting ensures the models are comparable for building a meta learner seed = 2023, fold_assignment = "Modulo", nfolds = 10, keep_cross_validation_predictions = TRUE) ####################################################### ### PREPARE ENSEMBLE MODEL ####################################################### ### get the models' IDs from the AutoML and grid searches. ### this is all that is needed before building the ensemble, ### i.e., to specify the model IDs that should be evaluated. ids <- c(h2o.get_ids(aml), h2o.get_ids(grid)) top <- ensemble(models = ids, training_frame = prostate, strategy = "top") search <- ensemble(models = ids, training_frame = prostate, strategy = "search") ####################################################### ### EVALUATE THE MODELS ####################################################### h2o.auc(aml@leader) # best model identified by h2o.automl h2o.auc(h2o.getModel(grid@model_ids[[1]])) # best model identified by grid search h2o.auc(top$model). # ensemble model with 'top' search strategy h2o.auc(search$model). # ensemble model with 'search' search strategy ## End(Not run)
Multiple trained H2O models are stacked to create an ensemble
ensemble( models, training_frame, newdata = NULL, family = "binary", strategy = c("search"), model_selection_criteria = c("auc", "aucpr", "mcc", "f2"), min_improvement = 1e-05, max = NULL, top_rank = seq(0.01, 0.99, 0.01), stop_rounds = 3, reset_stop_rounds = TRUE, stop_metric = "auc", seed = -1, verbatim = FALSE )ensemble( models, training_frame, newdata = NULL, family = "binary", strategy = c("search"), model_selection_criteria = c("auc", "aucpr", "mcc", "f2"), min_improvement = 1e-05, max = NULL, top_rank = seq(0.01, 0.99, 0.01), stop_rounds = 3, reset_stop_rounds = TRUE, stop_metric = "auc", seed = -1, verbatim = FALSE )
models |
H2O search grid or AutoML grid or a character vector of H2O model IDs.
the |
training_frame |
h2o training frame (data.frame) for model training |
newdata |
h2o frame (data.frame). the data.frame must be already uploaded on h2o server (cloud). when specified, this dataset will be used for evaluating the models. if not specified, model performance on the training dataset will be reported. |
family |
model family. currently only |
strategy |
character. the current available strategies are |
model_selection_criteria |
character, specifying the performance metrics that
should be taken into consideration for model selection. the default are
|
min_improvement |
numeric. specifies the minimum improvement in model evaluation metric to qualify further optimization search. |
max |
integer. specifies maximum number of models for each criteria to be extracted. the
default value is the |
top_rank |
numeric vector. specifies percentage of the top models taht
should be selected. if the strategy is |
stop_rounds |
integer. number of stoping rounds, in case the model stops improving |
reset_stop_rounds |
logical. if TRUE, every time the model improves the stopping rounds penalty is resets to 0. |
stop_metric |
character. model stopping metric. the default is |
seed |
random seed (recommended) |
verbatim |
logical. if TRUE, it reports additional information about the progress of the model training, particularly used for debugging. |
a list including the ensemble model and the top-rank models that were used in the model
E. F. Haghish
## Not run: # load the required libraries for building the base-learners and the ensemble models library(h2o) library(autoEnsemble) # initiate the h2o server h2o.init(ignore_config = TRUE, nthreads = 2, bind_to_localhost = FALSE, insecure = TRUE) # upload data to h2o cloud prostate_path <- system.file("extdata", "prostate.csv", package = "h2o") prostate <- h2o.importFile(path = prostate_path, header = TRUE) ### H2O provides 2 types of grid search for tuning the models, which are ### AutoML and Grid. Below, I tune 2 set of model grids and use them both ### for building the ensemble, just to set an example ... ####################################################### ### PREPARE AutoML Grid (takes a couple of minutes) ####################################################### # run AutoML to tune various models (GLM, GBM, XGBoost, DRF, DeepLearning) for 120 seconds y <- "CAPSULE" prostate[,y] <- as.factor(prostate[,y]) #convert to factor for classification aml <- h2o.automl(y = y, training_frame = prostate, max_runtime_secs = 120, include_algos=c("DRF","GLM", "XGBoost", "GBM", "DeepLearning"), # this setting ensures the models are comparable for building a meta learner seed = 2023, nfolds = 10, keep_cross_validation_predictions = TRUE) ####################################################### ### PREPARE H2O Grid (takes a couple of minutes) ####################################################### # make sure equal number of "nfolds" is specified for different grids grid <- h2o.grid(algorithm = "gbm", y = y, training_frame = prostate, hyper_params = list(ntrees = seq(1,50,1)), grid_id = "ensemble_grid", # this setting ensures the models are comparable for building a meta learner seed = 2023, fold_assignment = "Modulo", nfolds = 10, keep_cross_validation_predictions = TRUE) ####################################################### ### PREPARE ENSEMBLE MODEL ####################################################### ### get the models' IDs from the AutoML and grid searches. ### this is all that is needed before building the ensemble, ### i.e., to specify the model IDs that should be evaluated. ids <- c(h2o.get_ids(aml), h2o.get_ids(grid)) top <- ensemble(models = ids, training_frame = prostate, strategy = "top") search <- ensemble(models = ids, training_frame = prostate, strategy = "search") ####################################################### ### EVALUATE THE MODELS ####################################################### h2o.auc(aml@leader) # best model identified by h2o.automl h2o.auc(h2o.getModel(grid@model_ids[[1]])) # best model identified by grid search h2o.auc(top$model). # ensemble model with 'top' search strategy h2o.auc(search$model). # ensemble model with 'search' search strategy ## End(Not run)## Not run: # load the required libraries for building the base-learners and the ensemble models library(h2o) library(autoEnsemble) # initiate the h2o server h2o.init(ignore_config = TRUE, nthreads = 2, bind_to_localhost = FALSE, insecure = TRUE) # upload data to h2o cloud prostate_path <- system.file("extdata", "prostate.csv", package = "h2o") prostate <- h2o.importFile(path = prostate_path, header = TRUE) ### H2O provides 2 types of grid search for tuning the models, which are ### AutoML and Grid. Below, I tune 2 set of model grids and use them both ### for building the ensemble, just to set an example ... ####################################################### ### PREPARE AutoML Grid (takes a couple of minutes) ####################################################### # run AutoML to tune various models (GLM, GBM, XGBoost, DRF, DeepLearning) for 120 seconds y <- "CAPSULE" prostate[,y] <- as.factor(prostate[,y]) #convert to factor for classification aml <- h2o.automl(y = y, training_frame = prostate, max_runtime_secs = 120, include_algos=c("DRF","GLM", "XGBoost", "GBM", "DeepLearning"), # this setting ensures the models are comparable for building a meta learner seed = 2023, nfolds = 10, keep_cross_validation_predictions = TRUE) ####################################################### ### PREPARE H2O Grid (takes a couple of minutes) ####################################################### # make sure equal number of "nfolds" is specified for different grids grid <- h2o.grid(algorithm = "gbm", y = y, training_frame = prostate, hyper_params = list(ntrees = seq(1,50,1)), grid_id = "ensemble_grid", # this setting ensures the models are comparable for building a meta learner seed = 2023, fold_assignment = "Modulo", nfolds = 10, keep_cross_validation_predictions = TRUE) ####################################################### ### PREPARE ENSEMBLE MODEL ####################################################### ### get the models' IDs from the AutoML and grid searches. ### this is all that is needed before building the ensemble, ### i.e., to specify the model IDs that should be evaluated. ids <- c(h2o.get_ids(aml), h2o.get_ids(grid)) top <- ensemble(models = ids, training_frame = prostate, strategy = "top") search <- ensemble(models = ids, training_frame = prostate, strategy = "search") ####################################################### ### EVALUATE THE MODELS ####################################################### h2o.auc(aml@leader) # best model identified by h2o.automl h2o.auc(h2o.getModel(grid@model_ids[[1]])) # best model identified by grid search h2o.auc(top$model). # ensemble model with 'top' search strategy h2o.auc(search$model). # ensemble model with 'search' search strategy ## End(Not run)
Multiple model performance metrics are computed for each model
evaluate(id, newdata = NULL, ...)evaluate(id, newdata = NULL, ...)
id |
a character vector of H2O model IDs retrieved from H2O Grid search
or AutoML random search. the |
newdata |
h2o frame (data.frame). the data.frame must be already uploaded on h2o server (cloud). when specified, this dataset will be used for evaluating the models. if not specified, model performance on the training dataset will be reported. |
... |
arguments to be passed to |
a data.frame of various model performance metrics for each model
E. F. Haghish
## Not run: library(h2o) library(h2otools) #for h2o.get_ids() function library(autoEnsemble) # initiate the H2O server to train a grid of models h2o.init(ignore_config = TRUE, nthreads = 2, bind_to_localhost = FALSE, insecure = TRUE) # Run a grid search or AutoML search prostate_path <- system.file("extdata", "prostate.csv", package = "h2o") prostate <- h2o.importFile(path = prostate_path, header = TRUE) y <- "CAPSULE" prostate[,y] <- as.factor(prostate[,y]) #convert to factor for classification aml <- h2o.automl(y = y, training_frame = prostate, max_runtime_secs = 30, seed = 2023, nfolds = 10, keep_cross_validation_predictions = TRUE) # get the model IDs from the H2O Grid search or H2O AutoML Grid ids <- h2otools::h2o.get_ids(aml) # evaluate all the models and return a dataframe evals <- evaluate(id = ids) ## End(Not run)## Not run: library(h2o) library(h2otools) #for h2o.get_ids() function library(autoEnsemble) # initiate the H2O server to train a grid of models h2o.init(ignore_config = TRUE, nthreads = 2, bind_to_localhost = FALSE, insecure = TRUE) # Run a grid search or AutoML search prostate_path <- system.file("extdata", "prostate.csv", package = "h2o") prostate <- h2o.importFile(path = prostate_path, header = TRUE) y <- "CAPSULE" prostate[,y] <- as.factor(prostate[,y]) #convert to factor for classification aml <- h2o.automl(y = y, training_frame = prostate, max_runtime_secs = 30, seed = 2023, nfolds = 10, keep_cross_validation_predictions = TRUE) # get the model IDs from the H2O Grid search or H2O AutoML Grid ids <- h2otools::h2o.get_ids(aml) # evaluate all the models and return a dataframe evals <- evaluate(id = ids) ## End(Not run)
extracts the model IDs from H2O AutoML object or H2O grid
h2o.get_ids(automl)h2o.get_ids(automl)
automl |
a h2o |
a character vector of trained models' names (IDs)
E. F. Haghish
## Not run: library(h2o) library(autoEnsemble) h2o.init(ignore_config = TRUE, nthreads = 2, bind_to_localhost = FALSE, insecure = TRUE) prostate_path <- system.file("extdata", "prostate.csv", package = "h2o") prostate <- h2o.importFile(path = prostate_path, header = TRUE) y <- "CAPSULE" prostate[,y] <- as.factor(prostate[,y]) #convert to factor for classification aml <- h2o.automl(y = y, training_frame = prostate, max_runtime_secs = 30) # get the model IDs ids <- h2o.get_ids(aml) ## End(Not run)## Not run: library(h2o) library(autoEnsemble) h2o.init(ignore_config = TRUE, nthreads = 2, bind_to_localhost = FALSE, insecure = TRUE) prostate_path <- system.file("extdata", "prostate.csv", package = "h2o") prostate <- h2o.importFile(path = prostate_path, header = TRUE) y <- "CAPSULE" prostate[,y] <- as.factor(prostate[,y]) #convert to factor for classification aml <- h2o.automl(y = y, training_frame = prostate, max_runtime_secs = 30) # get the model IDs ids <- h2o.get_ids(aml) ## End(Not run)
Multiple model performance metrics are computed
modelSelection( eval, family = "binary", top_rank = 0.01, max = NULL, model_selection_criteria = c("auc", "aucpr", "mcc", "f2") )modelSelection( eval, family = "binary", top_rank = 0.01, max = NULL, model_selection_criteria = c("auc", "aucpr", "mcc", "f2") )
eval |
an object of class |
family |
model family. currently only |
top_rank |
numeric. what percentage of the top model should be selected? the default value is top 1% models. |
max |
integer. specifies maximum number of models for each criteria to be extracted. the
default value is the |
model_selection_criteria |
character, specifying the performance metrics that
should be taken into consideration for model selection. the default are
|
a matrix of F-Measures for different thresholds or the highest F-Measure value
E. F. Haghish
## Not run: library(h2o) library(h2otools) #for h2o.get_ids() function library(h2oEnsemble) # initiate the H2O server to train a grid of models h2o.init(ignore_config = TRUE, nthreads = 2, bind_to_localhost = FALSE, insecure = TRUE) # Run a grid search or AutoML search prostate_path <- system.file("extdata", "prostate.csv", package = "h2o") prostate <- h2o.importFile(path = prostate_path, header = TRUE) y <- "CAPSULE" prostate[,y] <- as.factor(prostate[,y]) #convert to factor for classification aml <- h2o.automl(y = y, training_frame = prostate, max_runtime_secs = 30, seed = 2023, nfolds = 10, keep_cross_validation_predictions = TRUE) # get the model IDs from the H2O Grid search or H2O AutoML Grid ids <- h2otools::h2o.get_ids(aml) # evaluate all the models and return a dataframe evals <- evaluate(id = ids) # perform model selection (up to top 10% of each criteria) select <- modelSelection(eval = evals, top_rank = 0.1)) ## End(Not run)## Not run: library(h2o) library(h2otools) #for h2o.get_ids() function library(h2oEnsemble) # initiate the H2O server to train a grid of models h2o.init(ignore_config = TRUE, nthreads = 2, bind_to_localhost = FALSE, insecure = TRUE) # Run a grid search or AutoML search prostate_path <- system.file("extdata", "prostate.csv", package = "h2o") prostate <- h2o.importFile(path = prostate_path, header = TRUE) y <- "CAPSULE" prostate[,y] <- as.factor(prostate[,y]) #convert to factor for classification aml <- h2o.automl(y = y, training_frame = prostate, max_runtime_secs = 30, seed = 2023, nfolds = 10, keep_cross_validation_predictions = TRUE) # get the model IDs from the H2O Grid search or H2O AutoML Grid ids <- h2otools::h2o.get_ids(aml) # evaluate all the models and return a dataframe evals <- evaluate(id = ids) # perform model selection (up to top 10% of each criteria) select <- modelSelection(eval = evals, top_rank = 0.1)) ## End(Not run)
Defines criteria for ending the optimization search
stopping_criteria( df, round, stop, min_improvement, stop_rounds = 3, reset_stop_rounds = TRUE, stop_metric = "auc" )stopping_criteria( df, round, stop, min_improvement, stop_rounds = 3, reset_stop_rounds = TRUE, stop_metric = "auc" )
df |
data.frame. includes the metrics of ensemblem model performance |
round |
integer. the current round of optimization |
stop |
integer. current round of stopping penalty |
min_improvement |
numeric. specifies the minimum improvement in model evaluation metric to qualify further optimization search. |
stop_rounds |
integer. number of stoping rounds, in case the model stops improving |
reset_stop_rounds |
logical. if TRUE, everytime the model improves the stopping rounds penalty is resets to 0. |
stop_metric |
character. model stopping metric. the default is |
a matrix of F-Measures for different thresholds or the highest F-Measure value
E. F. Haghish