In gabrielcrepeault/xgbmr: Implémentation en R d'un modèle de micro-réserve utilisant l'algorithme XGBoost.

TODO: texte pour accompagner le code présenté ici.

Importation des packages nécessaires et données

library(mlr)
library(tidyverse)
library(xgbmr)

## Importer les 3 données différents et créer les 3 tâches
traindt_a <- read.csv('../data/train_a.csv') # modèle A
traindt_b <- read.csv('../data/train_b.csv') # modèle B
traindt_c <- read.csv('../data/train_c.csv') # modèle C
tsk_a <- makeRegrTask(data = select(traindt_a, -ClNr), target = 'Ultimate')
tsk_b <- makeRegrTask(data = select(traindt_b, -ClNr), target = 'Ultimate')
tsk_c <- makeRegrTask(data = select(traindt_c, -ClNr), target = 'Ultimate')

## Créer les environnements d'entrainement (fonction custom de xgbmr)
lapply(c('modele_a', 'modele_b', 'modele_c'), setup_training_env)

## Créer le learner. Hyperparamètres fixés.
xgb_pars <- list(
    objective="reg:squarederror",
    eval_metric = 'rmse',
    eta = 0.1,
    max_depth = 3,
    min_child_weight = 20, 
    # gamma = 10,
    verbose = 0,
    subsample = 0.5,
    colsample_bytree = 0.5
    # alpha = 1,
    # lambda = 1
)
lrn_init <- makeLearner(cl = 'regr.xgboost', id = "xgb_initial", par.vals = xgb_pars)

## Seed pour résultats reproductibles
set.seed(2019)

## Modèle A ####
ntree_a <- tune_nrounds(lrn_init, task = tsk_a, root = "", export = F, plot = F,
                        early_stopping_rounds = 10, iters = 3000)
fit_a <- setLearnerId(learner = ntree_a$updated_learner, id = "modele_a") %>% 
    train(task = tsk_a)

saveRDS(fit_a, file = file.path('models', fit_a$learner$id, paste0(fit_a$learner$id, '.rda')))

## Modèle B ####
ntree_b <- tune_nrounds(lrn_init, task = tsk_b, root = "", export = F, plot = F,
                        early_stopping_rounds = 10, iters = 3000)
fit_b <- setLearnerId(learner = ntree_b$updated_learner, id = "modele_b") %>% 
    train(task = tsk_b)

saveRDS(fit_b, file = file.path('models', fit_b$learner$id, paste0(fit_b$learner$id, '.rda')))

## Modèle C ####
ntree_c <- tune_nrounds(lrn_init, task = tsk_c, root = "", export = F, plot = F,
                        early_stopping_rounds = 10, iters = 3000) 
fit_c <- setLearnerId(learner = ntree_c$updated_learner, id = "modele_c") %>% 
    train(task = tsk_a)

saveRDS(fit_c, file = file.path('models', fit_c$learner$id, paste0(fit_c$learner$id, '.rda')))

gabrielcrepeault/xgbmr documentation built on Dec. 30, 2019, 11:51 p.m.