ruvnet Learn
Train an RL agent on episode data to derive a policy.
When to use
- Repeated task type with measurable reward and 50+ episodes
- User asks to "train", "build a policy", "make it learn"
- Pre-deployment of an autonomous agent that should pick actions itself
Algorithms (the bandit picks the right one)
| Algo | Best for |
|---|---|
| Q-Learning | Tabular state-spaces, discrete actions |
| SARSA | On-policy variant, conservative exploration |
| DQN | High-dimensional state, neural Q-fn |
| PPO | Continuous control, high-dim action |
| Actor-Critic | Baseline reduction, stable training |
| Policy Gradient | Direct policy parameterization |
| Decision Transformer | Offline RL on trajectories |
| MCTS | Tree-search planning under known dynamics |
| Model-Based RL | Sample-efficient when env model is learnable |
If you don't know which to pick, call agentdb_learning_route first — the bandit suggests one based on past performance on similar task signatures.
API
agentdb_learning_train(
algorithm: <one of above> // or 'auto' to let bandit pick
episodes: [<episodeId>, ...] // or task name → fetch automatically
hyperparams: { lr, gamma, epsilon, ... }
iterations: N
)
After training
- The bandit logs the resulting reward distribution on the chosen algorithm.
- The trained model is stored in the AgentDB skill library as a callable skill.
- Future calls to
agentdb_learning_route(task)may pick this trained skill if it scores well.
Don't
- Don't train on < 50 episodes — high-variance models overfit.
- Don't train multiple algorithms in parallel "to see which wins" — that's the bandit's job, and parallel training pollutes the reward signal.
- Don't ignore the route output. If
agentdb_learning_routesays "no algorithm has > 0.6 expected reward on this task", the answer is to gather more episodes, not to force-train.