This is documentation special for DeepRL Component.

You can read more detail about DRL in below pages

Deep Reinforcement Learning Component

Workflow

Just quoting shortly, DRL is a way to train machine learning by doing random action, saving it into an experience buffer, and then taking a random sample of the best action as the neural network data train.

Reinforcement Learning (Training only)

Exploration Rate

The exploration rate governs how often the agent chooses a random action instead of relying on its learned policy. the value varies between 0-1, 1 means 100% and 0 means 0%.

it is recommend to use 20% - 30%

also recommended to start high and decrease as learning progresses.

Linear Exploration Rate

Reduce Exploration rate, as we are training the model. simply after we tried N experience, the Exploration rate will become 0.

ExplorationRate = (1 - (NowExperience/Linear) * OriginalExplorationRate;

-1 means, constant exploration rate.