Fix prediction from loaded pickle. (#4516)

include/xgboost/learner.h

@@ -119,7 +119,7 @@ class Learner : public rabit::Serializable {
                        bool pred_leaf = false,
                        bool pred_contribs = false,
                        bool approx_contribs = false,
-                       bool pred_interactions = false) const = 0;
+                       bool pred_interactions = false) = 0;
 
   /*!
    * \brief Set additional attribute to the Booster.

src/learner.cc

@@ -329,6 +329,7 @@ class LearnerImpl : public Learner {
         const std::string prefix = "SAVED_PARAM_";
         if (kv.first.find(prefix) == 0) {
           const std::string saved_param = kv.first.substr(prefix.length());
+          bool is_gpu_predictor = saved_param == "predictor" && kv.second == "gpu_predictor";
 #ifdef XGBOOST_USE_CUDA
           if (saved_param == "predictor" || saved_param == "n_gpus"
               || saved_param == "gpu_id") {
@@ -346,10 +347,14 @@ class LearnerImpl : public Learner {
               << "  * JVM packages:   bst.setParam(\""
               << saved_param << "\", [new value])";
           }
+#else
+          if (is_gpu_predictor) {
+            cfg_["predictor"] = "cpu_predictor";
+            kv.second = "cpu_predictor";
+          }
 #endif  // XGBOOST_USE_CUDA
           // NO visiable GPU on current environment
-          if (GPUSet::AllVisible().Size() == 0 &&
-              (saved_param == "predictor" && kv.second == "gpu_predictor")) {
+          if (is_gpu_predictor && GPUSet::AllVisible().Size() == 0) {
             cfg_["predictor"] = "cpu_predictor";
             kv.second = "cpu_predictor";
           }
@@ -543,7 +548,7 @@ class LearnerImpl : public Learner {
   void Predict(DMatrix* data, bool output_margin,
                HostDeviceVector<bst_float>* out_preds, unsigned ntree_limit,
                bool pred_leaf, bool pred_contribs, bool approx_contribs,
-               bool pred_interactions) const override {
+               bool pred_interactions) override {
     bool multiple_predictions = static_cast<int>(pred_leaf) +
                                 static_cast<int>(pred_interactions) +
                                 static_cast<int>(pred_contribs);
@@ -712,10 +717,11 @@ class LearnerImpl : public Learner {
    * \param ntree_limit limit number of trees used for boosted tree
    *   predictor, when it equals 0, this means we are using all the trees
    */
-  inline void PredictRaw(DMatrix* data, HostDeviceVector<bst_float>* out_preds,
-                         unsigned ntree_limit = 0) const {
+  void PredictRaw(DMatrix* data, HostDeviceVector<bst_float>* out_preds,
+                  unsigned ntree_limit = 0) {
     CHECK(gbm_ != nullptr)
         << "Predict must happen after Load or InitModel";
+    ConfigurationWithKnownData(data);
     gbm_->PredictBatch(data, out_preds, ntree_limit);
   }
 

tests/python-gpu/test_pickling.py

@@ -4,7 +4,9 @@ import unittest
 import numpy as np
 import subprocess
 import os
+import sys
 import xgboost as xgb
+from xgboost import XGBClassifier
 
 model_path = './model.pkl'
 
@@ -17,6 +19,17 @@ def build_dataset():
     return x, y
 
 
+def save_pickle(bst, path):
+    with open(path, 'wb') as fd:
+        pickle.dump(bst, fd)
+
+
+def load_pickle(path):
+    with open(path, 'rb') as fd:
+        bst = pickle.load(fd)
+    return bst
+
+
 class TestPickling(unittest.TestCase):
     def test_pickling(self):
         x, y = build_dataset()
@@ -27,8 +40,7 @@ class TestPickling(unittest.TestCase):
                  'verbosity': 1}
         bst = xgb.train(param, train_x)
 
-        with open(model_path, 'wb') as fd:
-            pickle.dump(bst, fd)
+        save_pickle(bst, model_path)
         args = ["pytest",
                 "--verbose",
                 "-s",
@@ -51,3 +63,30 @@ class TestPickling(unittest.TestCase):
         status = subprocess.call(command, env=env, shell=True)
         assert status == 0
         os.remove(model_path)
+
+    def test_predict_sklearn_pickle(self):
+        x, y = build_dataset()
+
+        kwargs = {'tree_method': 'gpu_hist',
+                  'predictor': 'gpu_predictor',
+                  'verbosity': 2,
+                  'objective': 'binary:logistic',
+                  'n_estimators': 10}
+
+        model = XGBClassifier(**kwargs)
+        model.fit(x, y)
+
+        save_pickle(model, "model.pkl")
+        del model
+
+        # load model
+        model: xgb.XGBClassifier = load_pickle("model.pkl")
+        os.remove("model.pkl")
+
+        gpu_pred = model.predict(x, output_margin=True)
+
+        # Switch to CPU predictor
+        bst = model.get_booster()
+        bst.set_param({'predictor': 'cpu_predictor'})
+        cpu_pred = model.predict(x, output_margin=True)
+        np.testing.assert_allclose(cpu_pred, gpu_pred, rtol=1e-5)