The following document specifies exactly how each question in Project 2 will be graded.
QUESTION 1 (3 points)
The autograder will check that your agent can rapidly clear the openClassic layout ten times with random ghosts. If your Pacman agent dies more than twice, or thrashes around infinitely (i.e. repeatedly moving back and forth between two positions, making no progress), or spends more than 35s on any single game, then you will get zero points.
The autograder computes the average score for the ten tries and if the average is greater than or equal to 1200, you get full credit. If the average is between 1000 and 1200, you get 2 points. Otherwise you get 1 point.
QUESTION 2 (5 points)
The autograder will check your agent on smallClassic layout. If your Pacman agent thrashes around infinitely or spends more than 30s on the layout, you will get zero points.
Your implementation of minimaxAgent will be compared to ours on the smallClassic layout with and without 'stop' variations. The autograder rewards full credit if your agent chooses exactly the same optimal moves as our agent.
The autograder checks your agent with the search depth parameter of 2. It first assigns full credit to your agent, then checks as follows:
If your agent searchs one step or three steps instead, the autograder will deduct two points.
Else if your agent only uses a partial search ply, for example one min layer for two or more ghosts, the autograder will deduct two points.
Else if your agent makes at least one mistake on choosing optimal actions, the autograder will deduct two points. It will deduct another point if your agent makes more than six mistakes.
QUESTION 3 (3 points)
The autograder will check your agent on smallClassic layout. If your Pacman agent thrashes around infinitely or spends more than 30s on the layout, you will get zero points.
Your implementation of the alpha-beta pruning agent will be compared to ours on the smallClassic layout with and without 'stop' variations. The autograder rewards full credit if your agent chooses exactly the same optimal moves as our agent.
The autograder checks your agent with the search depth parameter of 2. It firstly assigns full credit to your agent, then checks as follows:
If your agent searchs one step or three steps instead, the autograder will deduct two points.
Else if your agent only uses a partial search ply, for example one min layer for two or more ghosts, the autograder will deduct two points.
Else if your agent makes at least one mistake on choosing optimal actions, the autograder will deduct two points. It will deduct another point if your agent makes more than six mistakes.
QUESTION 4 (3 points)
The autograder will check your agent on smallClassic layout. If your Pacman agent thrashes around infinitely or spends more than 30s on the layout, you will get zero point.
Your implementation of ExpectimaxAgent will be compared to ours on the smallClassic layout with and without 'stop' variations. The autograder rewards full credit if your agent chooses exactly the same optimal moves as our agent.
The autograder checks your agent with the search depth parameter of 2. It firstly assigns full credit to your agent, then checks as follows:
If your agent searchs one step or three steps instead, the autograder will deduct two points.
Else if your agent only uses a partial search ply, for example one min layer for two or more ghosts, the autograder will deduct two points.
Else if your agent makes at least one mistake on choosing optimal actions, the autograder will deduct two points. It will deduct another point if your agent makes more than six mistakes.
QUESTION 5 (6 points)
The autograder evaluates your agent with depth 2 search on smallClassic layout for ten times with random ghosts. If your agent never wins or gets timeout for any of the ten rounds, the autograder will assign zero points. The autograder will add two points if your agent wins at least once. Your agent will get one more point if it wins greater than or equal to five times, in which case the autograder will add one additional point if the average win score is greater than or equal to 800.