Added #

• Documentation for DartRlAction constructor and value property
• Documentation for DartRlState constructor and value property
• Documentation for DartRlAgent constructor with parameter descriptions

Changed #

• Example structure: Restructured examples to follow Dart package layout guidelines
  • Moved frozen_lake_example.dart to example/frozen_lake/main.dart with its own pubspec.yaml
  • Moved grid_world_example.dart to example/grid_world/main.dart with its own pubspec.yaml
  • Each example now has its own subdirectory with pubspec.yaml that depends on the parent package
• Updated README.md to reflect new example structure and running instructions

Changed #

• Major refactor: Simplified package to focus on pure Dart implementation
• Renamed DartRLState to DartRlState for consistent naming convention
• Renamed DartRLAction to DartRlAction for consistent naming convention
• Renamed DartRLStateAction to DartRlStateAction for consistent naming convention
• Split state and action classes into separate files (state.dart, action.dart, state_action.dart)
• Replaced Equatable-based equality with manual operator == and hashCode implementations
• Updated package description to emphasize simplicity
• Simplified README with streamlined examples and documentation
• Updated version to 0.2.0-alpha.1

Removed #

• Flutter SDK dependency: Package is now pure Dart
• Dependencies: Removed collection, equatable, and flutter packages
• Dev dependencies: Removed pedantic package
• Flutter integration: Removed AgentNotifier and trainStream functionality
• Flutter directory: Removed entire lib/src/flutter/ directory
• Advanced features: Removed DecaySchedule, TrainingStats, and QTableSerializer classes
• Flutter-specific files: Removed agent_notifier.dart, agent_stream.dart
• Additional files: Removed decay_schedules.dart, serialization.dart, training_stats.dart
• Flutter documentation: Removed Flutter integration sections from README
• Flutter examples: Removed example/flutter_rl_demo/ directory

Notes #

This version represents a significant simplification of the package, focusing on core reinforcement learning algorithms for pure Dart applications. Flutter support and advanced features have been removed to reduce complexity and dependencies.

Added #

• Flutter SDK dependency for seamless Flutter integration
• collection package (^1.17.0) for enhanced data structures
• equatable package (^2.0.5) for value equality in state and action classes
• AgentNotifier class: ChangeNotifier wrapper for Flutter state management integration
• trainStream extension method for reactive UI updates with stream-based training
• Decay schedules: LinearDecaySchedule and ExponentialDecaySchedule for epsilon decay
• TrainingStats class for tracking episode-level training metrics
• AggregatedStats class for computing statistics across multiple episodes
• QTableSerializer for saving and loading trained Q-tables to/from disk
• Comprehensive Flutter integration documentation in README with examples
• Complete Flutter demo app in example/flutter_rl_demo/ with real-time visualization
• Examples for both stream-based training and ChangeNotifier pattern
• Documentation on decay schedules, training statistics, and model persistence

Features #

• Real-time training visualization for Flutter applications
• Stream-based training with reactive UI updates via trainStream
• Flutter state management integration through AgentNotifier (ChangeNotifier pattern)
• Compatible with Provider, Riverpod, and other Flutter state management solutions
• Configurable epsilon decay schedules (linear and exponential)
• Training statistics tracking with episode-level and aggregated metrics
• Q-table serialization for saving and loading trained agents
• Interactive Flutter example demonstrating real-time RL training visualization
• Support for training progress monitoring with episode, reward, steps, and epsilon tracking
• Non-blocking asynchronous training for smooth UI performance

Changed #

• Refactored State to DartRLState for improved naming consistency
• Refactored Action to DartRLAction for improved naming consistency
• Refactored StateAction to DartRLStateAction for improved naming consistency

Added #

• Initial alpha release of dart_rl package
• Q-Learning algorithm implementation (QLearningAgent)
• SARSA algorithm implementation (SarsaAgent)
• Expected-SARSA algorithm implementation (ExpectedSarsaAgent)
• Environment interface for creating custom RL environments
• Agent base class with epsilon-greedy exploration strategy
• State, Action, and StateAction classes for representing RL components
• StepResult class for environment step results
• Grid World example environment
• Frozen Lake example environment
• Comprehensive unit tests
• Documentation and README with usage examples

Features #

• Support for discrete state and action spaces
• Configurable learning rate (α), discount factor (γ), and epsilon (ε)
• Epsilon-greedy exploration with decay functionality
• Q-table access for inspection and debugging
• Training methods for single episodes and multiple episodes
• Compatible with both Dart and Flutter applications

Added #

• Initial release of dart_rl package
• Q-Learning algorithm implementation (QLearningAgent)
• SARSA algorithm implementation (SarsaAgent)
• Expected-SARSA algorithm implementation (ExpectedSarsaAgent)
• Environment interface for creating custom RL environments
• Agent base class with epsilon-greedy exploration strategy
• State, Action, and StateAction classes for representing RL components
• StepResult class for environment step results
• Grid World example environment
• Frozen Lake example environment
• Comprehensive unit tests
• Documentation and README with usage examples

Features #

• Support for discrete state and action spaces
• Configurable learning rate (?), discount factor (?), and epsilon (?)
• Epsilon-greedy exploration with decay functionality
• Q-table access for inspection and debugging
• Training methods for single episodes and multiple episodes
• Compatible with both Dart and Flutter applications