Practice · Praxis

Multiple choice Hard sarsaq-learningon-policyoff-policy ~100s

On a grid where the shortest route runs along the edge of the lava, two agents both act ε-greedily with ε = 0.1. SARSA learns a path that stays a cell away from the lava; Q-learning learns the edge-hugging shortest path. Why?

A Q-learning handles large negative rewards badly, so it underestimates how bad the lava is and keeps walking beside it B SARSA uses a smaller learning rate in dangerous states, so its values there change more slowly and stay more cautious C SARSA explores less than Q-learning does, so it rarely reaches the edge path and never discovers that it is shorter D SARSA's target includes its own random steps, so edge cells carry the risk of stumbling in; Q-learning's max ignores it

Hints are free to take and they are counted. A correct answer after two hints moves your mastery less than one without — which is the honest way to score it.