sagmaker reinforcement learning - ddqn

[troybvo]

I'm using dueling double deep q network for my project. I'm experiencing some problems with the model deployed from the endpoint giving different prediction than the evaluation step. What could have caused this? Also, the operation I used for output calculation is 'main_level/agent/main/online/network_0/dueling_q_values_head_0/output' type=Add`. Any help would be appreciated.