> ML_ALGORITHM // STATE-ACTION-REWARD-STATE-ACTION-SARSA_v1.0
SARSA (State-Action-Reward-State-Action)
On-policy temporal difference algorithm that updates Q-values using the actual next action taken by the current policy rather than the greedy maximum.
Value-Based Model-Free RLreinforcement-learninghigh-intrinsicmedium (1k-100k)
Back to All AlgorithmsComputational Complexity
Training Complexity:O(episodes * steps)
Inference Complexity:O(|A|) lookup
Hardware Profile
CPU Friendly:Yes
Requires GPU:No
Memory Footprint:low
Interpretability & Data
Interpretability Tier:high-intrinsic
Training Data Needs:medium (1k-100k)
Interpretability Assessment
Learned Q-values reflect the safety penalties of the active exploration policy.
Suitable Tasks & Supported Modalities
Suitable Tasks:
reinforcement learningsafe exploration
Supported Modalities:
tabular
Implementing Libraries
gymnasium
torchrl
Foundational Literature
Common Pitfalls & Warnings
- Converges to sub-optimal conservative paths in dangerous environments compared to Q-learning