CogitoNTNU/DeepTactics-Muzero
Feb 2025 – Apr 2025
3 Months active
Languages Used
PyTorchPythonShellText
Technical Skills
AIAlgorithm DesignAlgorithm DevelopmentAlgorithm ImplementationBackend DevelopmentData Structures
Ludvig Øvrevik
PROFILE
Ludvig Øvrevik
Developed and integrated a Monte Carlo Tree Search (MCTS) engine and MuZero-style training workflow for the CogitoNTNU/DeepTactics-Muzero repository, focusing on robust AI planning and reinforcement learning. Leveraged Python and PyTorch to implement core MCTS components, including PUCT-based scoring, node management, and exploration strategies, and wired these into the game environment for end-to-end gameplay. Enhanced the training pipeline with self-play, replay buffers, and Optuna-based hyperparameter optimization, while improving code clarity and maintainability through targeted refactoring and debugging. Upgraded SLURM-based orchestration scripts to streamline experiment management, improve resource allocation, and unify logging for reproducible, scalable AI research workflows.
Overall Statistics
Feature vs Bugs
50%Features
Repository Contributions
49Total
Bugs
9
Commits
49
Features
9
Lines of code
1,128
Activity Months3
Your Network
17 peopleSame Organization
@ntnu.no5
Work History
April 2025
2 Commits • 1 Features
Apr 1, 2025April 2025 monthly summary for CogitoNTNU/DeepTactics-Muzero: Delivered targeted fixes and logging enhancements to the MCTS-based MuZero workflow, improving training stability, reproducibility, and debugging efficiency. Key outcomes include correcting MCTS PUCT score calculation, standardizing and redirecting hyperparameter tuning logs, and upgrading the SLURM-based experiment orchestration for better traceability and resource usage. These efforts reduce debugging time, accelerate iteration on tactics strategies, and deliver clearer experiment telemetry across runs.
2 Commits • 1 Features
Apr 1, 2025April 2025 monthly summary for CogitoNTNU/DeepTactics-Muzero: Delivered targeted fixes and logging enhancements to the MCTS-based MuZero workflow, improving training stability, reproducibility, and debugging efficiency. Key outcomes include correcting MCTS PUCT score calculation, standardizing and redirecting hyperparameter tuning logs, and upgrading the SLURM-based experiment orchestration for better traceability and resource usage. These efforts reduce debugging time, accelerate iteration on tactics strategies, and deliver clearer experiment telemetry across runs.
April 2025
March 2025
28 Commits • 7 Features
Mar 1, 2025March 2025 performance highlights for CogitoNTNU/DeepTactics-Muzero: Delivered an end-to-end MuZero training loop with self-play scaffolding and initial network training integration; hardened MCTS and environment integration with improved action_space handling, MinMaxStats usage, and reliable reward propagation; improved environment initialization (render_mode) and parameter readability (action_space_size); modernized training stack toward a PyTorch-centric loss, while aligning optimizer usage and initializing SharedStorage/ReplayBuffer for stable data flows; introduced Optuna-based hyperparameter optimization with expanded configuration (td_steps and num_unroll_steps) and richer replay sampling; plus reliability enhancements including testing for SharedStorage, improved logging, batch bug fixes, and deployment readiness with more nodes and Slurm CPU core scaling.
March 2025
28 Commits • 7 Features
Mar 1, 2025March 2025 performance highlights for CogitoNTNU/DeepTactics-Muzero: Delivered an end-to-end MuZero training loop with self-play scaffolding and initial network training integration; hardened MCTS and environment integration with improved action_space handling, MinMaxStats usage, and reliable reward propagation; improved environment initialization (render_mode) and parameter readability (action_space_size); modernized training stack toward a PyTorch-centric loss, while aligning optimizer usage and initializing SharedStorage/ReplayBuffer for stable data flows; introduced Optuna-based hyperparameter optimization with expanded configuration (td_steps and num_unroll_steps) and richer replay sampling; plus reliability enhancements including testing for SharedStorage, improved logging, batch bug fixes, and deployment readiness with more nodes and Slurm CPU core scaling.
February 2025
19 Commits • 1 Features
Feb 1, 2025February 2025: Implemented and integrated a robust Monte Carlo Tree Search (MCTS) engine for DeepTactics-Muzero, delivering end-to-end AI planning and game play. Completed core components (PUCT-based scoring, Node management, action/history modeling, Dirichlet noise, softmax exploration, run loop, backpropagation) and wired them into the Game class for cohesive gameplay. Added supportive constructs (Action, ActionHistory, Player, MinMaxStats) and MCTS utilities (select_child, expand_node, main_mcts). Stabilized the pipeline with tensorized observations, corrected action history handling, and cleaned up network output (removed policy_tensor). This work enhances AI planning diversity, stability, and end-to-end gameplay, enabling stronger decision making and easier further MuZero-style improvements.
19 Commits • 1 Features
Feb 1, 2025February 2025: Implemented and integrated a robust Monte Carlo Tree Search (MCTS) engine for DeepTactics-Muzero, delivering end-to-end AI planning and game play. Completed core components (PUCT-based scoring, Node management, action/history modeling, Dirichlet noise, softmax exploration, run loop, backpropagation) and wired them into the Game class for cohesive gameplay. Added supportive constructs (Action, ActionHistory, Player, MinMaxStats) and MCTS utilities (select_child, expand_node, main_mcts). Stabilized the pipeline with tensorized observations, corrected action history handling, and cleaned up network output (removed policy_tensor). This work enhances AI planning diversity, stability, and end-to-end gameplay, enabling stronger decision making and easier further MuZero-style improvements.
February 2025
Activity
Loading activity data...
Quality Metrics
Correctness84.2%
Maintainability86.4%
Architecture80.6%
Performance72.8%
AI Usage21.4%
Skills & Technologies
Programming Languages
PyTorchPythonShellText
Technical Skills
AIAlgorithm DesignAlgorithm DevelopmentAlgorithm ImplementationAlgorithm OptimizationAlgorithm RefactoringBackend DevelopmentCode ClarityCode RefactoringConfiguration ManagementData StructuresDebuggingDeep LearningDependency ManagementEnvironment Configuration
Repositories Contributed To
Overview of all repositories you've contributed to across your timeline