Action-Value coverage estimation on RiverSwim mostly-right target policy T=50
0.523Q-Value Estimate (s=1, a=0)Model-based bootstrap with percentile CI
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Model-based bootstrap with percentile CIn=500, Nominal Coverage=50%, Episode Length (T)=50, Target Policy=mostly-right2026.05 | 0.523 | 0.526 | 0.518 | |
| Model-based bootstrap with pivot CIn=500, Nominal Coverage=50%, Episode Length (T)=50, Target Policy=mostly-right2026.05 | 0.507 | 0.508 | 0.483 | |
| Plug-in (CLT) CIn=500, Nominal Coverage=50%, Episode Length (T)=50, Target Policy=mostly-right2026.05 | 0.503 | 0.499 | 0.462 | |
| Model-based bootstrap with percentile CIn=1000, Nominal Coverage=50%, Episode Length (T)=50, Target Policy=mostly-right2026.05 | 0.495 | 0.497 | 0.525 | |
| Plug-in (CLT) CIn=1000, Nominal Coverage=50%, Episode Length (T)=50, Target Policy=mostly-right2026.05 | 0.487 | 0.488 | 0.491 | |
| Model-based bootstrap with pivot CIn=1000, Nominal Coverage=50%, Episode Length (T)=50, Target Policy=mostly-right2026.05 | 0.48 | 0.485 | 0.477 | |
| Episodic bootstrap with percentile CIn=1000, Nominal Coverage=50%, Episode Length (T)=50, Target Policy=mostly-right2026.05 | 0.474 | 0.473 | 0.484 | |
| Episodic bootstrap with percentile CIn=500, Nominal Coverage=50%, Episode Length (T)=50, Target Policy=mostly-right2026.05 | 0.47 | 0.465 | 0.447 | |
| Episodic bootstrap with pivot CIn=500, Nominal Coverage=50%, Episode Length (T)=50, Target Policy=mostly-right2026.05 | 0.465 | 0.453 | 0.411 | |
| Episodic bootstrap with pivot CIn=1000, Nominal Coverage=50%, Episode Length (T)=50, Target Policy=mostly-right2026.05 | 0.444 | 0.449 | 0.452 | |
| Model-based bootstrap with percentile CIn=100, Nominal Coverage=50%, Episode Length (T)=50, Target Policy=mostly-right2026.05 | 0.434 | 0.445 | 0.036 | |
| Model-based bootstrap with pivot CIn=100, Nominal Coverage=50%, Episode Length (T)=50, Target Policy=mostly-right2026.05 | 0.427 | 0.406 | 0.422 | |
| Plug-in (CLT) CIn=100, Nominal Coverage=50%, Episode Length (T)=50, Target Policy=mostly-right2026.05 | 0.332 | 0.324 | 0.208 | |
| Model-based bootstrap with percentile CIn=50, Nominal Coverage=50%, Episode Length (T)=50, Target Policy=mostly-right2026.05 | 0.304 | 0.31 | 0.005 | |
| Model-based bootstrap with pivot CIn=50, Nominal Coverage=50%, Episode Length (T)=50, Target Policy=mostly-right2026.05 | 0.271 | 0.224 | 0.232 | |
| Episodic bootstrap with percentile CIn=100, Nominal Coverage=50%, Episode Length (T)=50, Target Policy=mostly-right2026.05 | 0.24 | 0.237 | 0.077 | |
| Episodic bootstrap with pivot CIn=100, Nominal Coverage=50%, Episode Length (T)=50, Target Policy=mostly-right2026.05 | 0.229 | 0.205 | 0.309 | |
| Plug-in (CLT) CIn=50, Nominal Coverage=50%, Episode Length (T)=50, Target Policy=mostly-right2026.05 | 0.195 | 0.188 | 0.119 | |
| Episodic bootstrap with pivot CIn=50, Nominal Coverage=50%, Episode Length (T)=50, Target Policy=mostly-right2026.05 | 0 | 0 | 0 | |
| Episodic bootstrap with percentile CIn=50, Nominal Coverage=50%, Episode Length (T)=50, Target Policy=mostly-right2026.05 | 0 | 0 | 0 |