DQNExportModel
python
DQNExportModel(model: DQNModel, export_format: Literal['q_values', 'action'], action_masker: ActionMasker | None = None) -> NoneWrapper module for exporting a DQNModel to ONNX with a simplified output format.
Transforms the raw DQNModel output into either Q-values or a single action index, handling both categorical and non-categorical DQN variants via get_q_values().
When an action mask is provided, untrained actions are set to -1e9 before computing the argmax or returning Q-values. This ensures the exported model's behaviour matches training, where only masked actions are considered.
Initialize the DQNExportModel.
Parameters:
model(DQNModel) – The DQNModel to wrap for export.export_format(Literal['q_values', 'action']) – The output format -"q_values"for Q-values per action, or"action"for the greedy action index.action_masker(ActionMasker | None) – Optional ActionMasker to apply before action selection. When provided, the mask is baked into the ONNX graph.