Change binning behavior to be same as PR microsoft#2342.

btrotta · Sep 10, 2019 · 9ed04a3 · 9ed04a3
1 parent 23dbb29
commit 9ed04a3
Show file tree

Hide file tree

Showing 2 changed files with 10 additions and 6 deletions.
diff --git a/src/io/bin.cpp b/src/io/bin.cpp
@@ -186,17 +186,22 @@ namespace LightGBM {
       }
     }
 
-    // include zero bounds if possible
+    // include zero bounds and infinity bound
     if (max_bin == 2) {
       if (left_cnt == 0) {
         bin_upper_bound.push_back(kZeroThreshold);
       } else {
         bin_upper_bound.push_back(-kZeroThreshold);
       }
     } else if (max_bin >= 3) {
-      bin_upper_bound.push_back(-kZeroThreshold);
-      bin_upper_bound.push_back(kZeroThreshold);
+      if (left_cnt > 0) {
+        bin_upper_bound.push_back(-kZeroThreshold);
+      }
+      if (right_start >= 0) {
+        bin_upper_bound.push_back(kZeroThreshold);
+      }
     }
+    bin_upper_bound.push_back(std::numeric_limits<double>::infinity());
 
     // add forced bounds, excluding zeros since we have already added zero bounds
     int i = 0;
@@ -207,7 +212,6 @@ namespace LightGBM {
         ++i;
       }
     }
-    bin_upper_bound.push_back(std::numeric_limits<double>::infinity());
     int max_to_insert = max_bin - static_cast<int>(bin_upper_bound.size());
     int num_to_insert = std::min(max_to_insert, static_cast<int>(forced_upper_bounds.size()));
     if (num_to_insert > 0) {
@@ -239,7 +243,7 @@ namespace LightGBM {
     }
     bin_upper_bound.insert(bin_upper_bound.end(), bounds_to_add.begin(), bounds_to_add.end());
     std::stable_sort(bin_upper_bound.begin(), bin_upper_bound.end());
-    CHECK(bin_upper_bound.size() <= max_bin);
+    CHECK(bin_upper_bound.size() <= static_cast<size_t>(max_bin));
     return bin_upper_bound;
   }
 

diff --git a/tests/python_package_test/test_engine.py b/tests/python_package_test/test_engine.py
@@ -915,7 +915,7 @@ def test_max_bin_by_feature(self):
         }
         lgb_data = lgb.Dataset(X, label=y)
         est = lgb.train(params, lgb_data, num_boost_round=1)
-        self.assertEqual(len(np.unique(est.predict(X))), 99)
+        self.assertEqual(len(np.unique(est.predict(X))), 100)
         params['max_bin_by_feature'] = [2, 100]
         lgb_data = lgb.Dataset(X, label=y)
         est = lgb.train(params, lgb_data, num_boost_round=1)