Package 'kfold'

add_previous

Description

add_previous

Usage

add_previous(d, nlead = 12)

Arguments

d	with the variable of Q_obs
nlead	the number of leads to add

---

Stratified k-fold split

Description

Splits observation indices into kfold groups, ensuring each group receives a representative range of the target variable y.

Usage

chunk_stratified(y, kfold = 5, seed = 1)

Arguments

y	Numeric vector of target values used for stratified splitting.
kfold	Number of folds.
seed	Random seed (currently unused; seed is fixed internally).

Value

A named list of length kfold, each element an integer vector of row indices belonging to that fold.

---

Build lagged feature matrices for multiple lead times

Description

Build lagged feature matrices for multiple lead times

Usage

feature_leads(data_full, leads = 1:12)

Arguments

data_full	A data.table / data.frame with columns Q_obs, P, PET_Romanenko, and Q_sim.
leads	Integer vector of lead times (in time steps) to build features for.

Value

A named list (one element per lead) of lists with X (feature matrix) and Y (response matrix).

---

Compute GOF across multiple lead-time kfold objects

Description

Compute GOF across multiple lead-time kfold objects

Usage

GOT_list(list_kfold, list_test, ..., idcol = "lead")

Arguments

...	Ignored.
idcol	Column name for the lead-time id column.
objects	Named list of kfold objects (one per lead time).
ds_test	Named list of test datasets (one per lead time), each a list with X and Y.

Value

A data.table of GOF metrics with columns lead and mode.

---

kfold_calib

Description

Calibrate a model on a single train/validation split.

Usage

kfold_calib(X, Y, FUN = xgboost, index = NULL, ..., ratio_valid = 0.3)

Arguments

X	Feature matrix (rows = observations).
Y	Response matrix (rows = observations).
FUN	Model fitting function with signature FUN(x_train, y_train, ...).
index	Integer vector of validation row indices. If NULL, the first floor(n * ratio_valid) rows are used.
...	Additional arguments forwarded to FUN.
ratio_valid	Fraction of rows used as validation when index = NULL.

---

kfold machine learning

Description

kfold machine learning

Usage

kfold_ml(
  X,
  Y,
  kfold = 5,
  FUN,
  ...,
  fn_chunk = chunk_stratified,
  .progress = TRUE
)

kfold_rf(X, Y, kfold = 5, FUN = ranger, ntree = 500, importance = "none", ...)

kfold_xgboost(X, Y, kfold = 5, FUN = xgboost, nrounds = 500, ...)

kfold_lm(X, Y, kfold = 5, ...)

Arguments

X	Feature matrix (rows = observations).
Y	Response matrix (rows = observations).
kfold	Number of folds.
FUN	Model fitting function with signature FUN(x_train, y_train, ...).
...	Additional arguments forwarded to FUN.
fn_chunk	Fold-splitting function; defaults to chunk_stratified().
.progress	Show a progress bar during fold iteration.
ntree	Number of trees for kfold_rf().
importance	Variable importance mode, one of 'none', 'impurity', 'impurity_corrected', 'permutation'. The 'impurity' measure is the Gini index for classification, the variance of the responses for regression and the sum of test statistics (see splitrule) for survival.
nrounds	Number of boosting iterations / rounds. Note that the number of default boosting rounds here is not automatically tuned, and different problems will have vastly different optimal numbers of boosting rounds.

See Also

ranger::ranger(), xgboost::xgboost()

Examples

set.seed(1)
n <- 100 ; p <- 2
X <- matrix(rnorm(n * p), n, p) # no intercept!
y <- as.matrix(rnorm(n))

## kfold
r_lm  <- kfold_lm(X, y)
r_xgb <- kfold_xgboost(X, y)
# r_rf  <- kfold_rf(X, y)

## 70%-30% split
r = kfold_calib(X, y, ratio_valid = 0.7, nrounds=500, verbose=FALSE)
r$gof

---

GOF

Description

Good of fitting

Usage

NSE(yobs, ysim, w, ...)

GOF(yobs, ...)

## Default S3 method:
GOF(
  yobs,
  ysim,
  w = NULL,
  include.cv = FALSE,
  include.r = TRUE,
  ...,
  idcol = "kfold",
  mode = "test"
)

## S3 method for class 'kfold'
GOF(yobs, test = NULL, ...)

Arguments

yobs	Numeric vector, observations
ysim	Numeric vector, corresponding simulated values
w	Numeric vector, weights of every points. If w included, when calculating mean, Bias, MAE, RMSE and NSE, w will be taken into considered.
...	Ignored.
include.cv	If true, cv will be included.
include.r	If true, r and R2 will be included.
idcol	Column name for the id column when binding multi-column results.
mode	Label inserted into the mode column when computing multi-column GOF.
test	A list with X and Y for external testing. If NULL, GOF will be calculated

Value

• RMSE root mean square error

• NSE NASH coefficient

• MAE mean absolute error

• AI Agreement index (only good points (w == 1)) participate to calculate. See details in Zhang et al., (2015).

• Bias bias

• Bias_perc bias percentage

• n_sim number of valid obs

• cv Coefficient of variation

• R2 correlation of determination

• R pearson correlation

• pvalue pvalue of R

References

1. https://en.wikipedia.org/wiki/Coefficient_of_determination

2. https://en.wikipedia.org/wiki/Explained_sum_of_squares

3. https://en.wikipedia.org/wiki/Nash%E2%80%93Sutcliffe_model_efficiency_coefficient

4. Zhang Xiaoyang (2015), http://dx.doi.org/10.1016/j.rse.2014.10.012

Examples

yobs <- rnorm(100)
ysim <- yobs + rnorm(100) / 4
GOF(yobs, ysim)

---

predict for kfold object

Description

predict for kfold object

Usage

## S3 method for class 'kfold'
predict(object, newdata = NULL, ..., mode = "test")

Arguments

object	A kfold object returned by kfold_ml().
newdata	New feature matrix for prediction. Required when mode = "test".
...	Additional arguments forwarded to the underlying model's predict method.
mode	Prediction mode: "train" (in-sample, hold-out fold masked), "valid" (out-of-fold only), or "test" (full new data).

---

previous_tn

Description

Build a lagged matrix: column t is the original series, columns t-1, t-2, … are progressively shifted (lagged) copies.

Usage

previous_tn(x, n = 7, prefix = "", ...)

## Default S3 method:
previous_tn(x, n = 7, prefix = "", ...)

## S3 method for class 'data.frame'
previous_tn(x, n = 7, ...)

Arguments

x	Numeric vector (default method) or data.frame to lag.
n	Number of lags to create.
prefix	Character prefix prepended to each column name.
...	Ignored.

Examples

set.seed(1)
x <- rnorm(10)
previous_tn(x, 7, "R1_")
# data.frame
d = data.frame(x)
previous_tn(d)

---