From 7ce6fab8f81b2c39222fb7c5009ffd43f30a8985 Mon Sep 17 00:00:00 2001
From: Paolo Bonini <45519144+PaoloBnn@users.noreply.github.com>
Date: Fri, 20 Sep 2019 19:26:16 +0200
Subject: [PATCH] update models

Updated models xgboost and lightgbm to deal with overfitting
---
 DESCRIPTION       |  2 +-
 R/model_lighgbm.R |  5 +++--
 R/model_xgboost.R | 26 ++++++++++++--------------
 3 files changed, 16 insertions(+), 17 deletions(-)

diff --git a/DESCRIPTION b/DESCRIPTION
index a1d569b..6b5ecde 100644
--- a/DESCRIPTION
+++ b/DESCRIPTION
@@ -1,7 +1,7 @@
 Package: Retip
 Type: Package
 Title: Retention Time Prediction for Metabolomics
-Version: 0.5.3
+Version: 0.5.4
 Authors@R: c(
     person("Paolo", "Bonini", , "pb@ngalab.com", c("aut", "cre")),
     person("Tobias", "Kind", , "tkind@ucdavis.edu",role = "aut"),
diff --git a/R/model_lighgbm.R b/R/model_lighgbm.R
index e289150..f9d155f 100644
--- a/R/model_lighgbm.R
+++ b/R/model_lighgbm.R
@@ -33,7 +33,7 @@ valids <- list(test = dtest)
 params <- list(objective = "regression", metric = "rmse")
 
 # building cross validation model
-modelcv <- lightgbm::lgb.cv(params, dtrain, nrounds=5000,nfold = 10, valids,verbose = 1, early_stopping_rounds = 1000, record = TRUE, eval_freq = 1L,stratified = TRUE)
+modelcv <- lightgbm::lgb.cv(params, dtrain, nrounds=5000,nfold = 10, valids,verbose = 1, early_stopping_rounds = 1000, record = TRUE, eval_freq = 1L,stratified = TRUE,max_depth=4,max_leaf=20,max_bin=100)
 
 # select the best iter in cross validation
 best.iter <- modelcv$best_iter
@@ -42,10 +42,11 @@ best.iter <- modelcv$best_iter
 params <- list(objective = "regression_l2",metric = "rmse")
 
 # building final model
-model <- lightgbm::lgb.train(params, dtrain, nrounds=best.iter, valids,verbose = 0, early_stopping_rounds =1000, record = TRUE, eval_freq = 1L)
+model <- lightgbm::lgb.train(params, dtrain, nrounds=best.iter, valids,verbose = 0, early_stopping_rounds =1000, record = TRUE, eval_freq = 1L,max_depth=4,max_leaf=20,max_bin=100)
 
 print(paste0("End training"))
 
+
 return(model)
 
 }
diff --git a/R/model_xgboost.R b/R/model_xgboost.R
index fc82be4..5ba0d99 100644
--- a/R/model_xgboost.R
+++ b/R/model_xgboost.R
@@ -13,26 +13,24 @@ fit.xgboost <- function(x){
       cv.ctrl <-caret::trainControl(method = "cv",number = 10)
 
       # These are the tune grid parameters
-      xgb.grid <- base::expand.grid(nrounds=c(100,200,300,400,500,600,700),
-                              max_depth = c(5),
-                              eta = c(0.025,0.05),
-                              gamma = c(0.01),
-                              colsample_bytree = c(0.75),
-                              subsample = c(0.50),
-                              min_child_weight = c(0))
+      xgb.grid <- base::expand.grid(nrounds=c(300,400,500,600,700,800,1000),
+                                    max_depth = c(2,3,4,5),
+                                    eta = c(0.01,0.02),
+                                    gamma = c(1),
+                                    colsample_bytree = c(0.5),
+                                    subsample = c(0.5),
+                                    min_child_weight = c(10))
 
       print("Computing model Xgboost  ... Please wait ...")
 
       # Model training using the above parameters
       set.seed(101)
       model_xgb <-caret::train(RT ~.,
-                        data=x,
-                        method="xgbTree",
-                        metric = "RMSE",
-                        trControl=cv.ctrl,
-                        tuneGrid=xgb.grid,
-                        tuneLength = 14)
-
+                               data=x,
+                               method="xgbTree",
+                               metric = "RMSE",
+                               trControl=cv.ctrl,
+                               tuneGrid=xgb.grid)