Things to do for this project (Sort of prioritised order):
- Write tests for all relative functions. - Complete! 24/1/08
- Pass all tests. - Complete! 1/2/08
- Create a functioning agent, able to receive input and give output. - Complete! 1/2/08
- Modify the splitField method to new method. - Complete! 11/2/08
- Deal with the aftereffects of modification. - Complete! 15/2/08
- Implement smarter placement of pieces by using piece contours. - Complete! 1/3/08
- Modify reward storage to store with regards to variable states. - Complete! 21/2/08
- Test on multiple MDPs.
- Have the agent able to see if a rotation will hit the ceiling to check if it is a valid rotation. - Complete! Sort of… 16/3/08
- Try eligibility traces. - Complete! 24-3-08
- Try more TD learning.
- Implement bias towards lower states, finishing lines, and wall states. - Somewhat Complete.
- Implement afterstate evaluation for smarter play. – Work in progress
- Cull states if the stack is too close to the ceiling.
- Test various decision making methods (E-greedy values, Softmax temperatures/cooling rates).
- Run a proving run and compare how I stand.
- Jumble tests for better coverage of errors.
- Have the best agent! – Work in progress!
These will be marked off when completed. More will be added as I realise more problems also.
This entry was posted on Wednesday, March 26th, 2008 at 12:35 pm and is filed under Honours Project Progress. You can follow any responses to this entry through the RSS 2.0 feed. You can leave a response, or trackback from your own site.