Reinforcement Learning model always gives different output

Question

I am trying to build a reinforcement learning model for hardware capacity optimisation. The state of the model would input like CPU capacity utilisation, memory utilisation. The model is supposed to predict what should be the CPU, memory etc I need to provision for my environment. The model uses DQN at its core and reward mechanism is based on the current capacity used.

The challenge I have is that every time I run with same input state and reward mechanism, I am getting different combination of hardware to be provisioned.

Is it possible that RL might give different output for same set of inputs and reward?

Answer 1

Reinforcement learning in principle should not be the source of stochasiticy, i.e. given exactly the same inputs the dqn model and the rl agent should give exactly the same output.

The source of stochasticity in your setting may be the random seed. Make sure to set it once and share it with all libraries that have a stochastic aspect (i.e. network initialization, batch selection, your simulator if you are using one).