imitation-learning-bc-dagger Nikelroid · COMPLETED
Behavior cloning vs DAgger on MuJoCo continuous control. On Hopper-v4 behavior cloning collapses to 23% of expert return under compounding distribution shift; DAgger recovers 99.8%. Gaussian PyTorch policy with learned log-std, full relabelling loop, and a hyperparameter sweep harness.
github.com/Nikelroid/imitation-learning-bc-dagger · ★ 0 · Forks 0 · Size 2.8 MB
SUMMARY
Technologies 5
Scored 5
Observed 5
Practices 6
Evidence 6
Skips 0
COVERAGE
Analyzed 23 files · 1 commits · 0 API calls
TECHNOLOGIES & DEPTH
Markdown LANGUAGE Depth 70
1 files · PRODUCTION
Python LANGUAGE Depth 70
11 files · PRODUCTION
pip BUILD_TOOL Depth 80
1 files · CONFIGURATION
NumPy LIBRARY Depth 74
6 files · PRODUCTION
TypeScript LANGUAGE Depth 58
1 files · PRODUCTION
PRACTICES
documentation · observedautomated_tests · absentcontinuous_integration · absentcontainerization · absentlinting · absentformatting · absent
ACTIVITY & OWNERSHIP
First commit 2026-08-14
Last commit 2026-08-14
Active months 1
Commits 1