我正在使用以下参数训练xgboost

library(xgboost)
library(pdp)

default_param <- list(
          objective = "reg:linear",booster = "gbtree",eta = 0.01,gamma = 1,max_depth = 2,min_child_weight = 1,subsample = 0.7,colsample_bytree = 1)

xgb_mod <- xgb.train(data = dtrain,params = default_param,nrounds = 1E5,watchlist = list(train = dtrain,eval = dval),monotone_constraints = monotoneVar,early_stopping_rounds = 2000,verbose = 0)

下图显示了一个变量（灌溉）对我的反应（作物产量下降的趋势）的部分依赖性。尽管我希望产量随灌溉而增加，但xgboost拟合的功能更多是一个阶梯函数。为了使此功能更平滑，我需要调整哪些参数？