Skip to main content

> ML_ALGORITHM // DEEP-DETERMINISTIC-POLICY-GRADIENT-DDPG_v1.0

Deep Deterministic Policy Gradient (DDPG)

Actor-critic off-policy algorithm that operates over continuous action spaces by combining deterministic policy gradients with deep Q-learning.

Actor-Critic Continuous RLreinforcement-learningblack-boxlarge (>100k)
Back to All Algorithms
Computational Complexity
Training Complexity:O(steps * batch_size * (actor + critic)_passes)
Inference Complexity:O(actor_forward)
Hardware Profile
CPU Friendly:No
Requires GPU:Yes
Memory Footprint:moderate
Interpretability & Data
Interpretability Tier:black-box
Training Data Needs:large (>100k)

Interpretability Assessment

Actor outputs continuous real-valued action vectors directly evaluated by the critic.

Suitable Tasks & Supported Modalities

Suitable Tasks:
reinforcement learningcontinuous robotics control
Supported Modalities:
tabular

Implementing Libraries

stable-baselines3
torchrl
ray-rllib

Foundational Literature

Common Pitfalls & Warnings
  • Extremely sensitive to hyperparameter tuning and Ornstein-Uhlenbeck exploration noise decay
  • Severe Q-value overestimation leading to policy collapse (resolved by TD3)