all_models_fit: Fit all models on the entire dataset.

In statist-bhfz/stackeR: Easy model stacking

Description

Fit list of models on the entire dataset.

Usage

all_models_fit(data, target, models, model_params, model_args,
  preproc_funs)

Arguments

data	data.table with all input data.
target	Target variable name (character).
models	Named list of fit functions from tuneR package (xgb_fit, lgb_fit etc.)
model_params	List of data.table's with tunable model parameters.
model_args	List of unchangeable model parameters.
preproc_funs	List of preprocessing functions (one function per model) which takes data.table data+split as input and returns processed data.table with same target and split columns.

Value

List of model objects.

Examples

# Input data
dt <- as.data.table(mtcars)

# List of models
models <- list("xgboost" = xgb_fit, "catboost" = catboost_fit)

# Model parameters (turn off early stopping)
xgb_params <- data.table(
    max_depth = 6,
    eta = 0.025,
    colsample_bytree = 0.9,
    subsample = 0.8,
    gamma = 0,
    min_child_weight = 5,
    alpha = 0,
    lambda = 1
)
xgb_args <- list(
    nrounds = 50,
    booster = "gbtree",
    eval_metric = "rmse",
    objective = "reg:linear",
    verbose = 0
)

catboost_params <- data.table(
    iterations = 100,
    learning_rate = 0.05,
    depth = 8,
    loss_function = "RMSE",
    eval_metric = "RMSE",
    random_seed = 42,
    logging_level = "Silent"
)
catboost_args <- NULL

model_params <- list(xgb_params, catboost_params)
model_args <- list(xgb_args, catboost_args)

# Dumb preprocessing function
# Real function will contain imputation, feature engineering etc.
# with all statistics computed on train folds and applied to validation fold
preproc_fun_example <- function(data) return(data[])
# List of preprocessing fuctions for each model
preproc_funs <- list(preproc_fun_example, preproc_fun_example)

all_models <- all_models_fit(data = dt,
                             target = "hp",
                             models = models,
                             model_params = model_params,
                             model_args = model_args,
                             preproc_funs = preproc_funs)

statist-bhfz/stackeR documentation built on Aug. 7, 2019, 4:57 a.m.