How to Disable Attribute Pruning in ONNX Runtime for Performance Optimization of Ensemble Tree Regressor in C++

[uqfam]

When I run a Random Forest Regressor / TreeEnsembleRegressor model in ONNX Runtime in C++, it automatically prunes removable attributes every time I load the model into ORT. This process takes over two minutes, which is by far the biggest chuck of the runtime of the ORT session. Is there any way I can disable the attribute pruning to make my model more efficient?

Here is the corresponding message i get when running the onnx model:

2024-11-14 12:05:55.0911151 [I:onnxruntime:, session_state.cc:260 onnxruntime::SessionState::PruneRemovableAttributes] removed 12 removable attributes for node 'TreeEnsembleRegressor' ('TreeEnsembleRegressor'), among attributes: base_values, nodes_falsenodeids, nodes_featureids, nodes_hitrates, nodes_missing_value_tracks_true, nodes_modes, nodes_nodeids, nodes_treeids, nodes_truenodeids, nodes_values, target_ids, target_treeids, target_nodeids, target_weightsbase_values_as_tensor, nodes_hitrates_as_tensor, nodes_values_as_tensor, class_weights_as_tensor.