sklearn binary classifier: compiled and Python model discrepancies? about treelite HOT 7 CLOSED

import random
import numpy as np

from sklearn.ensemble import RandomForestClassifier

import treelite.gallery.sklearn
import treelite.runtime


# First, generate sample data
my_range = range(1, 100)
X_data = [[random.choice(my_range) for _ in range(125)] for _ in range(76)]
X_test = X_data.pop()
y_data = [random.choice([0, 1]) for _ in range(75)]

# Second, prepare a classifier and compile it
model_py = RandomForestClassifier(
    n_estimators=100,
    max_features=2,
    max_depth=None,
    min_samples_split=2,
    min_samples_leaf=5,
    bootstrap=True,
    n_jobs=-1
)
model_py.fit(X_data, y_data)

model_file = './compiled.so'
icompiler = treelite.gallery.sklearn.import_model(model_py)
icompiler.export_lib(toolchain='clang', libpath=model_file, verbose=True, params={'parallel_comp': 8})
model_tr = treelite.runtime.Predictor(model_file, verbose=True)

# Third, compare: py vs. tr
result_py = model_py.predict([X_test])[0]

batch = treelite.runtime.Batch.from_npy2d(np.array([X_test]))
result_tr = model_tr.predict(batch)

print(result_py, result_tr)
assert result_py == result_tr # I would expect that, but...