Optimality vs Approximation: good enough beats perfect | RL Ch.3 Ep.7

A perfect chart of every strait is for sale, inked one route at a time. There are 4^60 candidate plans, more than 10^36. How close to optimal is good enough? No winter is long enough for that. The perfect plan is an ideal, and most real agents cannot afford it. Two sailors spend one identical winter at 1:06. One inks only the routes actually sailed and revises as the winter goes. The other keeps chasing the complete chart. In spring, the first one has paid for the season and the second has not. That is the argument for approximation in one story. Close enough, on the routes you actually take, beats perfect, forever out of reach. Chapters 0:00 Every strait 0:37 The bet 1:06 Same winter 2:23 Call it 3:14 The ledgers 4:06 The budget is finite 5:33 Spring 6:47 Approximation Chapter 3, Part 7 of 7. The series follows Reinforcement Learning: An Introduction chapter by chapter. Chapter 3 playlist:    • Reinforcement Learning Chapter 3: Finite M...   Full series, in order:    • Reinforcement Learning: An Introduction, f...   Previous: Optimal Policy (RL Ch.3 Ep.6):    • Optimal Policy: read off in one greedy ste...   Based on ideas from Richard S. Sutton and Andrew G. Barto, Reinforcement Learning: An Introduction (2nd ed.), sections 3.7 (Optimality and Approximation). #ReinforcementLearning #MachineLearning #FunctionApproximation