f5e2cda68ae6035793686780c952d2083d98cf54,rllib/agents/sac/tests/test_sac.py,TestSAC,test_sac_loss_function,#TestSAC#,74

Before Change

                // Actually convert to torch tensors (by accessing everything).
                input_ = policy._lazy_tensor_dict(input_)
                input_ = {k: input_[k] for k in input_.keys()}
                log_alpha = policy.model.log_alpha.detach().numpy()[0]

            // Only run the expectation once, should be the same anyways
            // for all frameworks.

After Change

                // Actually convert to torch tensors (by accessing everything).
                input_ = policy._lazy_tensor_dict(input_)
                input_ = {k: input_[k] for k in input_.keys()}
                log_alpha = policy.model.log_alpha.detach().cpu().numpy()[0]

            // Only run the expectation once, should be the same anyways
            // for all frameworks.

Instances