fixed conflict

.appveyor.yml

@@ -53,5 +53,5 @@ test_script:
         if (!$?) { $host.SetShouldExit(-1) }
       }  # run all examples
   - cd %APPVEYOR_BUILD_FOLDER%\examples\python-guide\notebooks
-  - conda install -q -y -n test-env ipywidgets notebook
+  - conda install -q -y -n test-env ipywidgets notebook "tornado=5.1.1"
   - jupyter nbconvert --ExecutePreprocessor.timeout=180 --to notebook --execute --inplace *.ipynb  # run all notebooks

.ci/test.sh

@@ -139,6 +139,6 @@ matplotlib.use\(\"Agg\"\)\
     sed -i'.bak' 's/graph.render(view=True)/graph.render(view=False)/' plot_example.py
     for f in *.py; do python $f || exit -1; done  # run all examples
     cd $BUILD_DIRECTORY/examples/python-guide/notebooks
-    conda install -q -y -n $CONDA_ENV ipywidgets notebook
+    conda install -q -y -n $CONDA_ENV ipywidgets notebook "tornado=5.1.1"
     jupyter nbconvert --ExecutePreprocessor.timeout=180 --to notebook --execute --inplace *.ipynb || exit -1  # run all notebooks
 fi

.ci/test_windows.ps1

@@ -38,6 +38,6 @@ if ($env:TASK -eq "regular") {
     python $file ; Check-Output $?
   }  # run all examples
   cd $env:BUILD_SOURCESDIRECTORY/examples/python-guide/notebooks
-  conda install -q -y -n $env:CONDA_ENV ipywidgets notebook
+  conda install -q -y -n $env:CONDA_ENV ipywidgets notebook "tornado=5.1.1"
   jupyter nbconvert --ExecutePreprocessor.timeout=180 --to notebook --execute --inplace *.ipynb ; Check-Output $?  # run all notebooks
 }

.gitignore

@@ -393,3 +393,7 @@ lightgbm.Rcheck/
 
 # Files generated by aspell
 **/*.bak
+
+# GraphViz artifacts
+*.gv
+*.gv.*

examples/python-guide/advanced_example.py

@@ -163,6 +163,27 @@ def binary_error(preds, train_data):
 
 print('Finished 40 - 50 rounds with self-defined objective function and eval metric...')
 
+
+# another self-defined eval metric
+# f(preds: array, train_data: Dataset) -> name: string, eval_result: float, is_higher_better: bool
+# accuracy
+def accuracy(preds, train_data):
+    labels = train_data.get_label()
+    return 'accuracy', np.mean(labels == (preds > 0.5)), True
+
+
+gbm = lgb.train(params,
+                lgb_train,
+                num_boost_round=10,
+                init_model=gbm,
+                fobj=loglikelihood,
+                feval=lambda preds, train_data: [binary_error(preds, train_data),
+                                                 accuracy(preds, train_data)],
+                valid_sets=lgb_eval)
+
+print('Finished 50 - 60 rounds with self-defined objective function '
+      'and multiple self-defined eval metrics...')
+
 print('Starting a new training job...')
 
 

examples/python-guide/plot_example.py

@@ -50,10 +50,10 @@
 ax = lgb.plot_importance(gbm, max_num_features=10)
 plt.show()
 
-print('Plotting 84th tree...')  # one tree use categorical feature to split
-ax = lgb.plot_tree(gbm, tree_index=83, figsize=(20, 8), show_info=['split_gain'])
+print('Plotting 54th tree...')  # one tree use categorical feature to split
+ax = lgb.plot_tree(gbm, tree_index=53, figsize=(15, 15), show_info=['split_gain'])
 plt.show()
 
-print('Plotting 84th tree with graphviz...')
-graph = lgb.create_tree_digraph(gbm, tree_index=83, name='Tree84')
+print('Plotting 54th tree with graphviz...')
+graph = lgb.create_tree_digraph(gbm, tree_index=53, name='Tree54')
 graph.render(view=True)

examples/python-guide/sklearn_example.py

@@ -51,11 +51,27 @@ def rmsle(y_true, y_pred):
         eval_metric=rmsle,
         early_stopping_rounds=5)
 
+
+# another self-defined eval metric
+# f(y_true: array, y_pred: array) -> name: string, eval_result: float, is_higher_better: bool
+# Relative Absolute Error (RAE)
+def rae(y_true, y_pred):
+    return 'RAE', np.sum(np.abs(y_pred - y_true)) / np.sum(np.abs(np.mean(y_true) - y_true)), False
+
+
+print('Starting training with multiple custom eval functions...')
+# train
+gbm.fit(X_train, y_train,
+        eval_set=[(X_test, y_test)],
+        eval_metric=lambda y_true, y_pred: [rmsle(y_true, y_pred), rae(y_true, y_pred)],
+        early_stopping_rounds=5)
+
 print('Starting predicting...')
 # predict
 y_pred = gbm.predict(X_test, num_iteration=gbm.best_iteration_)
 # eval
 print('The rmsle of prediction is:', rmsle(y_test, y_pred)[1])
+print('The rae of prediction is:', rae(y_test, y_pred)[1])
 
 # other scikit-learn modules
 estimator = lgb.LGBMRegressor(num_leaves=31)

python-package/lightgbm/basic.py

@@ -2437,6 +2437,11 @@ def get_split_value_histogram(self, feature, bins=None, xgboost_style=False):
             The feature name or index the histogram is calculated for.
             If int, interpreted as index.
             If string, interpreted as name.
+
+            Note
+            ----
+            Categorical features are not supported.
+
         bins : int, string or None, optional (default=None)
             The maximum number of bins.
             If None, or int and > number of unique split values and ``xgboost_style=True``,
@@ -2464,7 +2469,10 @@ def add(root):
                 else:
                     split_feature = root['split_feature']
                 if split_feature == feature:
-                    values.append(root['threshold'])
+                    if isinstance(root['threshold'], string_type):
+                        raise LightGBMError('Cannot compute split value histogram for the categorical feature')
+                    else:
+                        values.append(root['threshold'])
                 add(root['left_child'])
                 add(root['right_child'])
 

swig/lightgbmlib.i

@@ -1,10 +1,12 @@
 /* lightgbmlib.i */
 %module lightgbmlib
 %ignore LGBM_BoosterSaveModelToString;
+%ignore LGBM_BoosterGetEvalNames;
 %{
 /* Includes the header in the wrapper code */
 #include "../include/LightGBM/export.h"
 #include "../include/LightGBM/utils/log.h"
+#include "../include/LightGBM/utils/common.h"
 #include "../include/LightGBM/c_api.h"
 %}
 
@@ -16,12 +18,32 @@
 
 %inline %{
   char * LGBM_BoosterSaveModelToStringSWIG(BoosterHandle handle,
-					   int start_iteration,
-					   int num_iteration,
-					   int64_t buffer_len,
-					   int64_t* out_len) {
+                                           int start_iteration,
+                                           int num_iteration,
+                                           int64_t buffer_len,
+                                           int64_t* out_len) {
     char* dst = new char[buffer_len];
     int result = LGBM_BoosterSaveModelToString(handle, start_iteration, num_iteration, buffer_len, out_len, dst);
+    // Reallocate to use larger length
+    if (*out_len > buffer_len) {
+      delete [] dst;
+      int64_t realloc_len = *out_len;
+      dst = new char[realloc_len];
+      result = LGBM_BoosterSaveModelToString(handle, start_iteration, num_iteration, realloc_len, out_len, dst);
+    }
+    if (result != 0) {
+      return nullptr;
+    }
+    return dst;
+  }
+
+  char ** LGBM_BoosterGetEvalNamesSWIG(BoosterHandle handle,
+                                       int eval_counts) {
+    char** dst = new char*[eval_counts];
+    for (int i = 0; i < eval_counts; ++i) {
+      dst[i] = new char[128];
+    }
+    int result = LGBM_BoosterGetEvalNames(handle, &eval_counts, dst);
     if (result != 0) {
       return nullptr;
     }
@@ -49,6 +71,7 @@
 %array_functions(float, floatArray)
 %array_functions(int, intArray)
 %array_functions(long, longArray)
+%array_functions(char *, stringArray)
 
 /* Custom pointer manipulation template */
 %define %pointer_manipulation(TYPE,NAME)

tests/python_package_test/test_engine.py

@@ -1246,17 +1246,17 @@ def test_model_size(self):
 
     def test_get_split_value_histogram(self):
         X, y = load_boston(True)
-        lgb_train = lgb.Dataset(X, y)
+        lgb_train = lgb.Dataset(X, y, categorical_feature=[2])
         gbm = lgb.train({'verbose': -1}, lgb_train, num_boost_round=20)
         # test XGBoost-style return value
         params = {'feature': 0, 'xgboost_style': True}
-        self.assertTupleEqual(gbm.get_split_value_histogram(**params).shape, (10, 2))
-        self.assertTupleEqual(gbm.get_split_value_histogram(bins=999, **params).shape, (10, 2))
+        self.assertTupleEqual(gbm.get_split_value_histogram(**params).shape, (9, 2))
+        self.assertTupleEqual(gbm.get_split_value_histogram(bins=999, **params).shape, (9, 2))
         self.assertTupleEqual(gbm.get_split_value_histogram(bins=-1, **params).shape, (1, 2))
         self.assertTupleEqual(gbm.get_split_value_histogram(bins=0, **params).shape, (1, 2))
         self.assertTupleEqual(gbm.get_split_value_histogram(bins=1, **params).shape, (1, 2))
         self.assertTupleEqual(gbm.get_split_value_histogram(bins=2, **params).shape, (2, 2))
-        self.assertTupleEqual(gbm.get_split_value_histogram(bins=6, **params).shape, (6, 2))
+        self.assertTupleEqual(gbm.get_split_value_histogram(bins=6, **params).shape, (5, 2))
         self.assertTupleEqual(gbm.get_split_value_histogram(bins=7, **params).shape, (6, 2))
         if lgb.compat.PANDAS_INSTALLED:
             np.testing.assert_almost_equal(
@@ -1278,8 +1278,8 @@ def test_get_split_value_histogram(self):
             )
         # test numpy-style return value
         hist, bins = gbm.get_split_value_histogram(0)
-        self.assertEqual(len(hist), 22)
-        self.assertEqual(len(bins), 23)
+        self.assertEqual(len(hist), 23)
+        self.assertEqual(len(bins), 24)
         hist, bins = gbm.get_split_value_histogram(0, bins=999)
         self.assertEqual(len(hist), 999)
         self.assertEqual(len(bins), 1000)
@@ -1317,6 +1317,9 @@ def test_get_split_value_histogram(self):
                 mask = hist_vals > 0
                 np.testing.assert_array_equal(hist_vals[mask], hist[:, 1])
                 np.testing.assert_almost_equal(bin_edges[1:][mask], hist[:, 0])
+        # test histogram is disabled for categorical features
+        self.assertRaises(lgb.basic.LightGBMError, gbm.get_split_value_histogram, 2)
+
 
     def test_early_stopping_for_only_first_metric(self):
         X, y = load_boston(True)