Package 'perpetual'

Train a PerpetualBooster model

Description

Perpetual is a self-generalizing gradient boosting machine that doesn't need hyperparameter optimization. It automatically finds the best configuration based on the provided budget.

Usage

perpetual(
  x,
  y,
  objective = "LogLoss",
  budget = NULL,
  iteration_limit = NULL,
  stopping_rounds = NULL,
  max_bin = NULL,
  num_threads = NULL,
  missing = NULL,
  allow_missing_splits = NULL,
  create_missing_branch = NULL,
  missing_node_treatment = NULL,
  log_iterations = NULL,
  quantile = NULL,
  reset = NULL,
  timeout = NULL,
  memory_limit = NULL,
  seed = NULL,
  calibration_method = NULL,
  save_node_stats = NULL,
  ...
)

Arguments

x	A matrix or data.frame of features.
y	A vector of targets (numeric for regression, factor/integer for classification).
objective	A string specifying the objective function. Default is "LogLoss".
budget	A numeric value ensuring the training time does not exceed this budget (in normalized units).
iteration_limit	An integer limit on the number of iterations.
stopping_rounds	An integer for early stopping.
max_bin	Integer, max number of bins for histograms.
num_threads	Integer, number of threads to use.
missing	Value to consider as missing data. Default is NaN.
allow_missing_splits	Boolean.
create_missing_branch	Boolean. Whether to create a separate branch for missing values (ternary trees).
missing_node_treatment	String. How to handle weights for missing nodes if create_missing_branch is True. Options: "None", "AssignToParent", "AverageLeafWeight", "AverageNodeWeight".
log_iterations	Integer.
quantile	Numeric.
reset	Boolean.
timeout	Numeric.
memory_limit	Numeric.
seed	Integer seed for reproducibility.
calibration_method	String specifying the calibration method for prediction intervals. Options: "WeightVariance", "MinMax", "GRP", "Conformal".
save_node_stats	Boolean. Whether to save node statistics (required for some calibration methods and importance types).
...	Additional arguments.

Value

A PerpetualBooster object.

---

Get the base score of the model

Description

Get the base score of the model

Usage

perpetual_base_score(model)

Arguments

model

A PerpetualBooster object.

Value

Numeric, base score.

---

Calibrate a PerpetualBooster model

Description

Calibrate a PerpetualBooster model

Usage

perpetual_calibrate(model, x, y, x_cal, y_cal, alpha, method = NULL)

Arguments

model	A PerpetualBooster object.
x	Validation features.
y	Validation targets.
x_cal	Calibration features.
y_cal	Calibration targets.
alpha	Calibration parameter.
method	String specifying the calibration method to use. If NULL, uses the method configured in the model.

---

Get feature importance

Description

Get feature importance

Usage

perpetual_importance(model, method = "gain", normalize = TRUE)

Arguments

model	A PerpetualBooster object.
method	String, method for importance (e.g. "gain").
normalize	Boolean.

Value

A named vector of importances.

---

Load a PerpetualBooster model

Description

Load a PerpetualBooster model

Usage

perpetual_load(path)

Arguments

path	String, path to the saved model.

Value

A PerpetualBooster object.

---

Get the number of trees in the model

Description

Get the number of trees in the model

Usage

perpetual_n_trees(model)

Arguments

model

A PerpetualBooster object.

Value

Integer, number of trees.

---

Save a PerpetualBooster model

Description

Save a PerpetualBooster model

Usage

perpetual_save(model, path)

Arguments

model	A PerpetualBooster object.
path	String, path to save the model.

---

Dump model to JSON string

Description

Dump model to JSON string

Usage

perpetual_to_json(model)

Arguments

model

A PerpetualBooster object.

Value

String, JSON representation of the model.

---

PerpetualBooster

Description

PerpetualBooster object for creating and loading models.

Usage

PerpetualBooster

Format

An object of class list of length 2.

---

Predict using a PerpetualBooster model

Description

Predict using a PerpetualBooster model

Usage

## S3 method for class 'PerpetualBooster'
predict(
  object,
  newdata,
  type = c("class", "prob", "raw", "contribution", "interval"),
  method = "average",
  ...
)

Arguments

object	A PerpetualBooster object.
newdata	A matrix or data.frame of new data to predict on.
type	Type of prediction: "class", "prob", "raw", "contribution", or "interval".
method	Method for prediction contributions. Default "average".
...	Additional arguments.

Value

A vector or matrix of predictions.

---

Print PerpetualBooster

Description

Print PerpetualBooster

Usage

## S3 method for class 'PerpetualBooster'
print(x, ...)

Arguments

x	A PerpetualBooster object.
...	Additional arguments.

---