lgb.train_with_cv: lgb.train_with_cv

In 6chaoran/suw: utility functions for streamlining underwriting models

Description

Train LightGBM with cross-validation

Usage

lgb.train_with_cv(
  df.train,
  df.valid,
  fnames,
  label,
  fnames.cat,
  id = NULL,
  rules = NULL,
  params = NULL,
  cv.verbose = 0,
  train.verbose = 1,
  cv = FALSE,
  nfold = 5,
  score.fun = Metrics::auc,
  ...
)

Arguments

df.train	data.frame for train set
df.valid	data.frame for valid set
fnames	all feature names
label	label name
fnames.cat	categorical feature names
id	colnames of id, if not provide, default to use rowname
rules	rules to index categorical features
params	list of params for lightgbm
cv.verbose	whether verbose when cv
train.verbose	whether verbose when train
cv	whether perform a k-fold cross validation, default at FALSE
nfold	number of folds of cross validation, if cv is TRUE
score.fun	score functino for validation, defaults to auc
...	other parameters in lgb.train

Value

list of (model, rules, fnames)

Examples

label <- 'label'
fnames <- c('Sex','Class','Age','Freq')
fnames.cat <- c('Sex','Age','Class')
df <- data.frame(Titanic)
df$label <- ifelse(df$Survived == "Yes", 1 ,0)
in.train <- runif(nrow(df)) < 0.8
df.train <- df[in.train, ]
df.valid <- df[!in.train, ]
bst <- lgb.train_with_cv(df.train, df.valid,
                       fnames = fnames,
                       label = label,
                       fnames.cat = fnames.cat,
                       rules = NULL, cv = TRUE,
                       num_leaves = 63,
                       learning_rate = 1.0,
                       nrounds = 10L,
                       min_data = 1L,
                       cv.verbose = 1,
                       train.verbose = 1,
                       eval = 'auc',
                       eval_freq = 10,
                       nfold = 2L,
                       early_stopping_rounds = 5L,
                       objective = "binary")
lgb.save_model(bst, './saved_model','lgb_baseline', verbose = TRUE)
bst.loaded <- lgb.load_model('./saved_model','lgb_baseline')
preds <- lgb.predict(bst.loaded, df.valid)

6chaoran/suw documentation built on July 29, 2020, 5:31 p.m.