def input_schema(self) -> LossInputSchema:
return LossInputSchema(head_outputs=['new_policy_probs'],
agent_inputs=[
'actions', 'old_policy_probs',
'clip_param_rescaler'
],
targets=['advantages'])
def input_schema(self) -> LossInputSchema:
return LossInputSchema(head_outputs=['pred'],
agent_inputs=[],
targets=['target'])
def input_schema(self) -> LossInputSchema:
return LossInputSchema(head_outputs=['new_policy_values'],
agent_inputs=['old_policy_values'],
targets=['target_values'])