The easiest way to test the performance of a model currently is to play it against a random agent. This is not necessarily indicative of a better model since only challenging it would yield a good model.
Instead, it would be good to analytically understand if the agent performs well in particular circumstances. For certain states, there is an action that guarantees a win. It would be good to get a value indicating if the agent picks these particular actions across many scenarios.
The easiest way to test the performance of a model currently is to play it against a random agent. This is not necessarily indicative of a better model since only challenging it would yield a good model.
Instead, it would be good to analytically understand if the agent performs well in particular circumstances. For certain states, there is an action that guarantees a win. It would be good to get a value indicating if the agent picks these particular actions across many scenarios.