| Type: | Package |
| Title: | Structural Forest for the Heterogeneous Newsvendor Model |
| Version: | 0.1.0 |
| Description: | Implements the structural forest methodology for the heterogeneous newsvendor model. The package provides tools to prepare data, fit honest newsvendor trees and forests, and obtain point and distributional predictions for demand decisions under uncertainty. |
| URL: | https://github.com/MurphyLiCN/SFHNV |
| License: | MIT + file LICENSE |
| Depends: | R (≥ 3.6) |
| Imports: | stats |
| Suggests: | future.apply, testthat (≥ 3.0.0) |
| Encoding: | UTF-8 |
| Config/testthat/edition: | 3 |
| RoxygenNote: | 7.3.3 |
| NeedsCompilation: | no |
| Packaged: | 2025-09-22 01:56:31 UTC; murphy |
| Author: | Mengfei Li [aut, cre] |
| Maintainer: | Mengfei Li <murphy.li.cn@gmail.com> |
| Repository: | CRAN |
| Date/Publication: | 2025-09-30 08:50:14 UTC |
Fit a Structural Forest Heterogeneous Newsvendor tree
Description
Builds an honest tree that estimates the structural parameter of the heterogeneous newsvendor model using the SFHNV algorithm.
Usage
NW_Tree(
data,
honest_ratio = 1,
min_size = 50,
max_depth = 50,
features = NULL,
approximate = FALSE,
max_candidates = 256,
leaf_round_digits = 1L
)
Arguments
data |
A data frame or output from |
honest_ratio |
Ratio of the estimation subsample to the splitting subsample. |
min_size |
Minimum number of observations in each child node. |
max_depth |
Maximum depth of the tree. |
features |
Optional subset of features (names or indices) to consider at each split. |
approximate |
Logical; if |
max_candidates |
Maximum candidate split points per feature when |
leaf_round_digits |
Control the rounding of demand samples when fitting leaf CDFs. Use negative values to disable rounding. |
Value
A list representing the fitted tree.
Examples
data <- data.frame(x1 = rnorm(200), x2 = rnorm(200), D = rnorm(200), Q = rnorm(200))
tree <- NW_Tree(data, min_size = 20, max_depth = 5)
preds <- predict_tree(tree, data)
Prepare data for SFHNV trees and forests
Description
Converts a data frame with outcome quantities into a numeric matrix representation used by the Structural Forest for the Heterogeneous Newsvendor (SFHNV) estimators.
Usage
NW_prepare(data)
Arguments
data |
A |
Value
A list with prepared matrices (X), outcomes (D, Q), binary indicators (z),
feature names, and the dimensions n and p.
Examples
data <- data.frame(x1 = rnorm(100), x2 = rnorm(100), D = rnorm(100), Q = rnorm(100))
prep <- NW_prepare(data)
str(prep)
Internal helper utilities
Description
These functions support the Structural Forest for the Heterogeneous Newsvendor estimators and are not exported.
Fit an SFHNV random forest
Description
Fit an SFHNV random forest
Usage
build_random_forest(
data,
honest_ratio = 1,
min_size = 50,
max_depth = 50,
num_trees = 100,
feature_choose = "sqrt",
parallel = TRUE,
approximate = FALSE,
max_candidates = 256,
leaf_round_digits = 1L
)
Arguments
data |
A data frame or output from |
honest_ratio |
Ratio of the estimation subsample to the splitting subsample. |
min_size |
Minimum number of observations in each child node. |
max_depth |
Maximum depth of each tree. |
num_trees |
Number of trees to build. |
feature_choose |
Strategy for selecting features at each split. One of
|
parallel |
Logical; if |
approximate |
Logical; if |
max_candidates |
Maximum candidate split points per feature when |
leaf_round_digits |
Rounding control for demand samples in leaf CDF estimation. |
Value
A list of SFHNV trees.
Examples
data <- data.frame(x1 = rnorm(200), x2 = rnorm(200), D = rnorm(200), Q = rnorm(200))
forest <- build_random_forest(data, num_trees = 5, min_size = 20)
Predict SFHNV random forest conditional CDF values
Description
Predict SFHNV random forest conditional CDF values
Usage
predict_cdf_forest(
forest,
observations,
d_values,
parallel = TRUE,
agg = "mean",
trim_prop = 0.05
)
Arguments
forest |
A list of trees produced by |
observations |
Data frame of new observations. |
d_values |
Scalar or vector of demand thresholds. |
parallel |
Logical; if |
agg |
Aggregation strategy across trees ( |
trim_prop |
Trimming proportion when |
Value
Numeric vector of CDF estimates.
Examples
data <- data.frame(x1 = rnorm(100), x2 = rnorm(100), D = rnorm(100), Q = rnorm(100))
forest <- build_random_forest(data, num_trees = 3, min_size = 15)
predict_cdf_forest(forest, data, d_values = 0)
Predict conditional CDF values from an SFHNV tree
Description
Predict conditional CDF values from an SFHNV tree
Usage
predict_cdf_tree(tree, observations, d_values)
Arguments
tree |
An object produced by |
observations |
Data frame of new observations containing the same features as the training data. |
d_values |
Either a scalar demand threshold applied to all observations, or a numeric vector with one value per observation. |
Value
Numeric vector of CDF values.
Examples
data <- data.frame(x = rnorm(50), D = rnorm(50), Q = rnorm(50))
tree <- NW_Tree(data, min_size = 10, max_depth = 3)
predict_cdf_tree(tree, data, d_values = 0)
Predict SFHNV random forest point estimates
Description
Predict SFHNV random forest point estimates
Usage
predict_forest(forest, observations, trim_prop = 0.05, parallel = TRUE)
Arguments
forest |
A list of trees produced by |
observations |
Data frame of new observations. |
trim_prop |
Optional trimming proportion used in the robust aggregation. |
parallel |
Logical; if |
Value
Numeric vector of aggregated predictions.
Examples
data <- data.frame(x1 = rnorm(100), x2 = rnorm(100), D = rnorm(100), Q = rnorm(100))
forest <- build_random_forest(data, num_trees = 3, min_size = 15)
predict_forest(forest, data)
Predict SFHNV tree point estimates
Description
Predict SFHNV tree point estimates
Usage
predict_tree(tree, observations)
Arguments
tree |
An object produced by |
observations |
Data frame of new observations containing the same features as the training data. |
Value
Numeric vector of predicted structural parameters.
Examples
data <- data.frame(x = rnorm(50), D = rnorm(50), Q = rnorm(50))
tree <- NW_Tree(data, min_size = 10, max_depth = 3)
predict_tree(tree, data)