Add multiclass test with larger n_classes

tests/test_tree_output.py

@@ -3,6 +3,7 @@
 import pytest
 from hypothesis import given, settings
 from hypothesis import strategies as st
+from sklearn.datasets import make_classification
 
 import lleaves
 
@@ -120,3 +121,37 @@ def test_forest_llvm_mode_cat(data, llvm_lgbm_model_cat):
     np.testing.assert_array_almost_equal(
         llvm_model.predict([input_data]), lgbm_model.predict([input_data]), decimal=15
     )
+
+
+def test_multiclass_generated(tmpdir):
+    """Check prediction equality on a freshly trained multiclass model"""
+    X, y = make_classification(
+        n_samples=5000,
+        n_features=20,
+        n_informative=7,
+        n_redundant=7,
+        n_classes=10,
+        random_state=1337,
+    )
+    d_train = lightgbm.Dataset(X, label=y)
+    params = {
+        "boosting_type": "gbdt",
+        "objective": "multiclass",
+        "metric": "multi_logloss",
+        "num_class": 10,
+    }
+    # will result in 3*10 trees
+    clf = lightgbm.train(params, d_train, 3)
+
+    model_file = str(tmpdir / "model.txt")
+    clf.save_model(model_file)
+
+    lgbm = lightgbm.Booster(model_file=model_file)
+    llvm = lleaves.Model(model_file=model_file)
+    llvm.compile()
+
+    # check predictions equal on the whole dataset
+    np.testing.assert_almost_equal(
+        lgbm.predict(X, n_jobs=2), llvm.predict(X, n_jobs=2), decimal=10
+    )
+    assert lgbm.num_model_per_iteration() == llvm.num_model_per_iteration()