def test_predict_no_preprocessor_no_training():
with tempfile.TemporaryDirectory() as tmpdir:
checkpoint = LightGBMCheckpoint.from_model(booster=model, path=tmpdir)
predictor = LightGBMPredictor.from_checkpoint(checkpoint)
data_batch = np.array([[1, 2], [3, 4], [5, 6]])
predictions = predictor.predict(data_batch)
assert len(predictions) == 3
def test_init():
preprocessor = DummyPreprocessor()
preprocessor.attr = 1
predictor = LightGBMPredictor(model=model, preprocessor=preprocessor)
with tempfile.TemporaryDirectory() as tmpdir:
# This somewhat convoluted procedure is the same as in the
# Trainers. The reason for saving model to disk instead
# of directly to the dict as bytes is due to all callbacks
# following save to disk logic. GBDT models are small
# enough that IO should not be an issue.
model.save_model(os.path.join(tmpdir, MODEL_KEY))
save_preprocessor_to_dir(preprocessor, tmpdir)
checkpoint = Checkpoint.from_directory(tmpdir)
checkpoint_predictor = LightGBMPredictor.from_checkpoint(checkpoint)
assert get_num_trees(checkpoint_predictor.model) == get_num_trees(
predictor.model)
assert checkpoint_predictor.preprocessor.attr == predictor.preprocessor.attr