tensorlayer/TensorLayer

Features

• Backend-Agnostic Deep Learning - Provides a backend-agnostic tensor library allowing model logic to run across different deep learning engines without code changes.
• Reinforcement Learning - Provides a toolkit for implementing reinforcement learning agents that learn optimal behaviors through environment interaction.
• Actor-Critic Architectures - Provides architectures that combine policy-based agents with value-based evaluators to stabilize training convergence.
• Deep Learning Architectures - Provides high-level layers and model abstractions to construct multi-layered neural network architectures.
• Experience Replay Buffers - Includes memory buffers that store past state transitions to break temporal correlations during reinforcement learning training.
• Neural Network Layers - Organizes neural networks using pre-defined architectural building blocks like dense, convolutional, or recurrent layers.
• Construction Abstractions - Ships high-level construction abstractions and reusable components to simplify the definition of neural network architectures.
• Neural Network Construction - Allows users to design deep learning architectures using high-level layers and model abstractions.
• Backend-Agnostic Engines - Provides a computational engine that decouples neural network operations from specific hardware backends for cross-platform execution.
• Policy Gradient Implementations - Implements architectures for optimizing decision-making policies in both discrete and continuous action spaces.
• Reinforcement Learning Value Estimators - Provides algorithms for iteratively updating action-value estimates to determine optimal policies in dynamic environments.
• Framework Abstraction Layers - Implements a framework abstraction layer that allows the same model logic to run across different deep learning backends.
• Neural Network Module Abstractions - Provides high-level class abstractions for encapsulating neural network layers and managing their associated parameters.
• Abstraction Layers - Provides a neural network abstraction layer to simplify the design of architectures across diverse deep learning engines.
• Deep Reinforcement Learning Implementations - Offers tools for developing deep reinforcement learning agents using neural networks, experience replay, and policy optimization.
• Priority-Based Replay Buffers - Provides a specialized memory buffer to sample high-importance state transitions for more efficient reinforcement learning.
• Policy and Value Function Approximators - Implements neural network architectures specifically designed to estimate values and action probabilities in reinforcement learning.
• Policy Gradient Optimizers - Provides gradient-based methods for updating policy parameters to optimize decision-making in discrete and continuous action spaces.
• Environment Wrappers - Provides wrappers that convert diverse external simulation interfaces into a uniform format for consistent agent interaction.
• RL Environment Wrapping Tools - Provides utilities to wrap third-party simulation environments into standardized formats compatible with RL workflows.
• Deep Learning Ecosystems - Deep learning and reinforcement learning library.
• Deep Learning Frameworks - Deep learning library built on TensorFlow.
• Machine Learning Frameworks - Deep learning and reinforcement learning library for TensorFlow.
• Frameworks and Libraries - Deep learning and reinforcement learning library for research.

Star history