Train Q-networks with Russian Doll MCTS in scored discrete-action environments.
npx skills add https://github.com/thistleknot/skills --skill deep-q-rl
Please help me install this Agent Skill. Skill: deep-q-rl Source: https://github.com/thistleknot/skills/tree/main/deep-q-rl Command: npx skills add https://github.com/thistleknot/skills --skill deep-q-rl
This Skill turns scored, discrete-action decision problems into an efficient training loop by learning a value function while using a progressive, search-guided policy improvement strategy.
evaluate(state) correlate instead of relying only on sparse terminal rewards.Use it for environments like board games, turn-based strategy, or any simulation where you can enumerate discrete actions, encode state tensors, and compute a current-player-perspective scalar score that correlates with ultimate success.
Use the deep-q-rl skill to train an agent by implementing the ScoredEnvironment interface with encode_state, evaluate, legal_actions, apply, and is_terminal for your environment, then run self-play or rollout-based training with Russian Doll MCTS and AHA enabled for training.
High-intent search queries and answers about installing and using this skill.
Deep Q-learning with Russian Doll MCTS trains policies from dense state scores by building a Q-style value head from a per-state evaluate correlate, using a target network and replay buffer to optimize decisions without waiting for sparse terminal rewards.
Russian Doll MCTS applies progressive narrowing search to wide discrete action spaces by using action categorization with weighted sampling and evaluating search leaves with a value-head Q-network, annealing funnel widths and exploration as the value function becomes more reliable.
Yes, self-play training works for turn-based strategy games or board games where you can encode state tensors, enumerate discrete legal actions, compute a current-player-perspective scalar score, and simulate transitions with an apply function.
AHA mistake correction detects evaluation drops after a chosen action during training and applies immediate corrective replay signal, improving policy decisions by penalizing moves that cause immediate state score degradation during self-play rollouts.
Yes, a bounded and calibrated evaluate score correlate is required to train the value-head Q-network with Bellman targets, ensuring the scalar per current-player POV accurately correlates with ultimate success for progressive narrowing search optimization.
Implement the ScoredEnvironment interface with encode_state, evaluate, legal_actions, apply, and is_terminal functions for your environment, then run self-play or rollout-based training with Russian Doll MCTS and AHA mistake correction enabled for progressive policy improvement.
PostHogEvaluate PostHog feature flags in Python applications with posthog APIs.
alijilani-devGuide Kubernetes deployment planning through structured Q&A scenario extraction.
GonglitianSelect actions between RL and BC policies using a dual-Q critic.
monktuiDevelop reinforcement learning plugins from logged agent experience using AgentDB.