h2o_automl: Automated H2O's AutoML

Description Usage Arguments Value List of algorithms Methods See Also Examples

View source: R/model_functions.R

Description

This function lets the user create a robust and fast model, using H2O's AutoML function. The result is a list with the best model, its parameters, datasets, performance metrics, variables importance, and plots. If the input is categorical, classification models will be trained and if is a continuous variable, regression models will be trained.

Usage

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
h2o_automl(
  df,
  y = "tag",
  ignore = NULL,
  train_test = NA,
  split = 0.7,
  weight = NULL,
  target = "auto",
  balance = FALSE,
  impute = FALSE,
  no_outliers = TRUE,
  unique_train = TRUE,
  center = FALSE,
  scale = FALSE,
  thresh = 10,
  seed = 0,
  nfolds = 5,
  max_models = 3,
  max_time = 10 * 60,
  start_clean = FALSE,
  exclude_algos = c("StackedEnsemble", "DeepLearning"),
  include_algos = NULL,
  plots = TRUE,
  alarm = TRUE,
  quiet = FALSE,
  print = TRUE,
  save = FALSE,
  subdir = NA,
  project = "ML Project",
  ...
)

## S3 method for class 'h2o_automl'
plot(x, ...)

## S3 method for class 'h2o_automl'
print(x, importance = TRUE, ...)

Arguments

df

Dataframe. Dataframe containing all your data, including the independent variable labeled as 'tag'. If you want to define which variable should be used instead, use the y parameter.

y

Variable or Character. Name of the independent variable.

ignore

Character vector. Force columns for the model to ignore

train_test

Character. If needed, df's column name with 'test' and 'train' values to split

split

Numeric. Value between 0 and 1 to split as train/test datasets. Value is for training set. Set value to 1 to train with all available data and test with same data (cross-validation will still be used when training). If train_test is set, value will be overwritten with its real split rate.

weight

Column with observation weights. Giving some observation a weight of zero is equivalent to excluding it from the dataset; giving an observation a relative weight of 2 is equivalent to repeating that row twice. Negative weights are not allowed.

target

Value. Which is your target positive value? If set to 'auto', the target with largest mean(score) will be selected. Change the value to overwrite. Only used when binary categorical model.

balance

Boolean. Auto-balance train dataset with under-sampling?

impute

Boolean. Fill NA values with MICE?

no_outliers

Boolean/Numeric. Remove y's outliers from the dataset? Will remove those values that are farther than n standard deviations from the independent variable's mean (Z-score). Set to TRUE for default (3) or numeric to set a different multiplier.

unique_train

Boolean. Keep only unique row observations for training data?

center, scale

Boolean. Using the base function scale, do you wish to center and/or scale all numerical values?

thresh

Integer. Threshold for selecting binary or regression models: this number is the threshold of unique values we should have in 'tag' (more than: regression; less than: classification)

seed

Integer. Set a seed for reproducibility. AutoML can only guarantee reproducibility if max_models is used because max_time is resource limited.

nfolds

Number of folds for k-fold cross-validation. Defaults to 5. Use 0 to disable cross-validation; this will also disable Stacked Ensemble (thus decreasing the overall model performance).

max_models, max_time

Numeric. Max number of models and seconds you wish for the function to iterate. Note that max_models guarantees reproducibility and max_time not (because it depends entirely on your machine's computational characteristics)

start_clean

Boolean. Erase everything in the current h2o instance before we start to train models? You may want to keep other models or not. To group results into a custom common AutoML project, you may use project_name argument.

exclude_algos, include_algos

Vector of character strings. Algorithms to skip or include during the model-building phase. Set NULL to ignore. When both are defined, only include_algos will be valid.

plots

Boolean. Create plots objects?

alarm

Boolean. Ping (sound) when done. Requires beepr.

quiet

Boolean. Quiet all messages, warnings, recommendations?

print

Boolean. Print summary when process ends?

save

Boolean. Do you wish to save/export results into your working directory?

subdir

Character. In which directory do you wish to save the results? Working directory as default.

project

Character. Your project's name

...

Additional parameters on h2o::h2o.automl

x

h2o_automl object

importance

Boolean. Print important variables?

Value

List. Trained model, predicted scores and datasets used, performance metrics, parameters, importance data.frame, seed, and plots when plots=TRUE.

List of algorithms

DRF

Distributed Random Forest, including Random Forest (RF) and Extremely-Randomized Trees (XRT)

GLM

Generalized Linear Model

XGBoost

eXtreme Grading Boosting

GBM

Gradient Boosting Machine

DeepLearning

Fully-connected multi-layer artificial neural network

StackedEnsemble

Stacked Ensemble

Read more here.

Methods

print

Use print method to print models stats and summary

plot

Use plot method to plot results using mplot_full()

See Also

Other Machine Learning: ROC(), conf_mat(), export_results(), gain_lift(), h2o_predict_API(), h2o_predict_MOJO(), h2o_predict_binary(), h2o_predict_model(), h2o_selectmodel(), impute(), iter_seeds(), lasso_vars(), model_metrics(), model_preprocess(), msplit()

Examples

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
## Not run: 
data(dft) # Titanic dataset
dft <- subset(dft, select = -c(Ticket, PassengerId, Cabin))

# Classification: Binomial - 2 Classes
r <- h2o_automl(dft, y = Survived, max_models = 1, impute = FALSE, target = "TRUE")

# Let's see all the stuff we have inside:
lapply(r, names)

# Classification: Multi-Categorical - 3 Classes
r <- h2o_automl(dft, Pclass, ignore = c("Fare", "Cabin"), max_time = 30, plots = FALSE)

# Regression: Continuous Values
r <- h2o_automl(dft, y = "Fare", ignore = c("Pclass"), exclude_algos = NULL, quiet = TRUE)
print(r)

# WITH PRE-DEFINED TRAIN/TEST DATAFRAMES
splits <- msplit(dft, size = 0.8)
splits$train$split <- "train"
splits$test$split <- "test"
df <- rbind(splits$train, splits$test)
r <- h2o_automl(df, "Survived", max_models = 1, train_test = "split")

## End(Not run)

lares documentation built on June 9, 2021, 9:06 a.m.