It is useful to navigate through a virtual environment.
Environment state \(S\), possible actions \(A\), and a value of each state/action \(Q\). Taking an action will impact the \(Q\) value.
$$ Q^{n e w}\left(s_{t}, a_{t}\right) \leftarrow \underbrace{Q\left(s_{t}, a_{t}\right)}_{\text {old value }}+\underbrace{\alpha}_{\text {learning rate }} \cdot \overbrace{(\underbrace{r_{t}}_{\text {reward }}+\underbrace{\gamma}_{\text {discount factor }} \cdot \underbrace{\max _{a} Q\left(s_{t+1}, a\right)}_{\text {estimate of optimal future value}}-\underbrace{Q\left(s_{t}, a_{t}\right)}_{\text {old value }})}_{\text {new value (temporal difference target) }}^{\text {temporal difference}} $$ aka. Markov Decision Process
SageMaker uses Tensorflow and MXNet
Input
Details
Hyperparameters
Depends on your task
Instance Choice
multi-core and multi-instance