There needs to be an interface defined for easily creating new training environments or 'Gyms' for the networks to learn in. These should have the ability to be completely separated from the networks. Data about the state should be generated, and then input for the next state is given. Those should be the only links between the two.
There needs to be an interface defined for easily creating new training environments or 'Gyms' for the networks to learn in. These should have the ability to be completely separated from the networks. Data about the state should be generated, and then input for the next state is given. Those should be the only links between the two.