randomforest_fit: Determine model fit of random forest

In ing-bank/tsforecast: Time Series Forecasting Pipeline

Description

randomforest_fit A function to gauge the fit of a model run of a random forest, given parameters. This is a function that is used to fine-tune the random forest when forecasting

Usage

randomforest_fit(ML_data, mtry, nodesize, ntree)

Arguments

ML_data	Dataset that has been prepared to run through randomForest. If originally a time series object, then it has gone through the decompose_ts_object_for_ML function and the first difference of the column of interest has been taken
mtry	randomForest parameter. The number of variables that are randomly sampled as candidates at each split. Default values are different for classification (sqrt(p)) and regression (p/3) where p is number of variables
nodesize	randomForest parameter. It is the minimum size of terminal nodes. Setting this numebr larger causes smaller trees to be grown and thus takes less time. Default values are different for classification (1) and regression (5)
ntree	randomForest parameter. It is the number of trees to grow. It should not be too small in order to eliminate any possible over-fitting and have each observation used at least a couple of times (randomForest default = 500)

Value

The mean absolute prediction error (MAPE), in percentage terms, of the model run

Examples

ML_data <- tstools::initialize_ts_forecast_data(
   data = dummy_gasprice, 
      date_col = "year_month", 
      col_of_interest = "gasprice", 
      group_cols = c("state", "oil_company"), 
      xreg_cols = c("spotprice", "gemprice")
   ) %>% 
   dplyr::filter(grouping == "state = New York   &   oil_company = CompanyA") %>% 
   tstools::transform_data_to_ts_object() %>% 
   decompose_ts_object_for_ML() %>% 
   dplyr::mutate(col_of_interest = col_of_interest - dplyr::lag(col_of_interest)) %>% 
   dplyr::filter(!is.na(col_of_interest))
randomforest_fit(
   ML_data = ML_data, 
   mtry = 8, 
   nodesize = 5, 
   ntree = 1000
)

ing-bank/tsforecast documentation built on Sept. 18, 2020, 9:40 a.m.