Fix the NaN recall in xgboost training: Update the tag of xgboost to v2.0.3; Add train code in train iteration; Passed all examples.

Author: cyccbxhl

examples/basic/src/main.rs

@@ -66,15 +66,15 @@ fn main() {
 
     // save and load model file
     println!("\nSaving and loading Booster model...");
-    booster.save("xgb.model").unwrap();
-    let booster = Booster::load("xgb.model").unwrap();
+    booster.save("xgb.json").unwrap();
+    let booster = Booster::load("xgb.json").unwrap();
     let preds2 = booster.predict(&dtest).unwrap();
     assert_eq!(preds, preds2);
 
     // save and load data matrix file
     println!("\nSaving and loading matrix data...");
     dtest.save("test.dmat").unwrap();
-    let dtest2 = DMatrix::load("test.dmat").unwrap();
+    let dtest2 = DMatrix::load_binary("test.dmat").unwrap();
     assert_eq!(booster.predict(&dtest2).unwrap(), preds);
 
     // error handling example

examples/generalised_linear_model/src/main.rs

@@ -12,8 +12,8 @@ fn main() {
 
     // load train and test matrices from text files (in LibSVM format)
     println!("Custom objective example...");
-    let dtrain = DMatrix::load("../../xgboost-sys/xgboost/demo/data/agaricus.txt.train").unwrap();
-    let dtest = DMatrix::load("../../xgboost-sys/xgboost/demo/data/agaricus.txt.test").unwrap();
+    let dtrain = DMatrix::load(r#"{"uri": "../../xgboost-sys/xgboost/demo/data/agaricus.txt.train?format=libsvm"}"#).unwrap();
+    let dtest = DMatrix::load(r#"{"uri": "../../xgboost-sys/xgboost/demo/data/agaricus.txt.test?format=libsvm"}"#).unwrap();
 
     // configure objectives, metrics, etc.
     let learning_params = parameters::learning::LearningTaskParametersBuilder::default()

src/booster.rs

@@ -148,8 +148,29 @@ impl Booster {
             dmats
         };
 
-        let bst = Booster::new_with_cached_dmats(&params.booster_params, &cached_dmats)?;
-        for i in 0..params.boost_rounds as i32 {
+        let mut bst = Booster::new_with_cached_dmats(&params.booster_params, &cached_dmats)?;
+        // load distributed code checkpoint from rabit
+        let mut version = bst.load_rabit_checkpoint()?;
+        debug!("Loaded Rabit checkpoint: version={}", version);
+        assert!(unsafe { xgboost_sys::RabitGetWorldSize() != 1 || version == 0 });
+        let start_iteration = version / 2;
+        for i in start_iteration..params.boost_rounds as i32 {
+            // distributed code: need to resume to this point
+            // skip first update if a recovery step
+            if version % 2 == 0 {
+                if let Some(objective_fn) = params.custom_objective_fn {
+                    debug!("Boosting in round: {}", i);
+                    bst.update_custom(params.dtrain, objective_fn)?;
+                } else {
+                    debug!("Updating in round: {}", i);
+                    bst.update(params.dtrain, i)?;
+                }
+                let _ = bst.save_rabit_checkpoint()?;
+                version += 1;
+            }
+
+            assert!(unsafe { xgboost_sys::RabitGetWorldSize() == 1 || version == xgboost_sys::RabitVersionNumber() });
+
             if let Some(eval_sets) = params.evaluation_sets {
                 let mut dmat_eval_results = bst.eval_set(eval_sets, i)?;
 
@@ -182,6 +203,10 @@ impl Booster {
                 }
                 println!();
             }
+
+            // do checkpoint after evaluation, in case evaluation also updates booster.
+            let _ = bst.save_rabit_checkpoint();
+            version += 1;
         }
 
         Ok(bst)
@@ -536,6 +561,16 @@ impl Booster {
         Ok(out_vec.join("\n"))
     }
 
+    pub(crate) fn load_rabit_checkpoint(&self) -> XGBResult<i32> {
+        let mut version = 0;
+        xgb_call!(xgboost_sys::XGBoosterLoadRabitCheckpoint(self.handle, &mut version))?;
+        Ok(version)
+    }
+
+    pub(crate) fn save_rabit_checkpoint(&self) -> XGBResult<()> {
+        xgb_call!(xgboost_sys::XGBoosterSaveRabitCheckpoint(self.handle))
+    }
+
     pub fn set_param(&mut self, name: &str, value: &str) -> XGBResult<()> {
         let name = ffi::CString::new(name).unwrap();
         let value = ffi::CString::new(value).unwrap();

src/parameters/learning.rs

@@ -81,11 +81,11 @@ impl Clone for Objective {
 impl ToString for Objective {
     fn to_string(&self) -> String {
         match *self {
-            Objective::RegLinear => "reg:linear".to_owned(),
+            Objective::RegLinear => "reg:squarederror".to_owned(),
             Objective::RegLogistic => "reg:logistic".to_owned(),
             Objective::BinaryLogistic => "binary:logistic".to_owned(),
             Objective::BinaryLogisticRaw => "binary:logitraw".to_owned(),
-            Objective::GpuRegLinear => "gpu:reg:linear".to_owned(),
+            Objective::GpuRegLinear => "gpu:reg:squarederror".to_owned(),
             Objective::GpuRegLogistic => "gpu:reg:logistic".to_owned(),
             Objective::GpuBinaryLogistic => "gpu:binary:logistic".to_owned(),
             Objective::GpuBinaryLogisticRaw => "gpu:binary:logitraw".to_owned(),

xgboost-sys/build.rs

@@ -33,7 +33,7 @@ fn main() {
     #[cfg(not(feature = "cuda"))]
     let mut dst = Config::new(&xgb_root);
 
-    let mut dst = dst.uses_cxx11()
+    let dst = dst.uses_cxx11()
         .define("BUILD_STATIC_LIB", "ON");
 
     #[cfg(target_os = "macos")]
@@ -54,7 +54,6 @@ fn main() {
 
     let bindings = bindgen::Builder::default()
         .header("wrapper.h")
-        .blocklist_item("std::.*")// stdlib is not well supported by bindgen
         .clang_args(&["-x", "c++", "-std=c++11"])
         .clang_arg(format!("-I{}", xgb_root.join("include").display()))
         .clang_arg(format!("-I{}", xgb_root.join("rabit/include").display()))
@@ -86,6 +85,8 @@ fn main() {
         println!("cargo:rustc-link-lib=c++");
         println!("cargo:rustc-link-lib=dylib=omp");
     } else {
+        println!("cargo:rustc-cxxflags=-std=c++17");
+        println!("cargo:rustc-link-lib=stdc++fs");
         println!("cargo:rustc-link-lib=stdc++");
         println!("cargo:rustc-link-lib=dylib=gomp");
     }