kaushik94 / angrybirds Goto Github PK

Hmmnnn gotta read some real MDP shit and try to implement it here.

• reinforcement : come up with a way that updates previous experiences much better without over-fitting.
• Math : how to model the environment more mathematically ? assign weights ? will that respect the structures properly ?
• causality : can probability theory be applied ? How probable is this action to cause this effect kind of questions ?
• prototyping : why should we do all this in the first place ? prototype ? how ?

This should evaluate the potential weak points of a structure so that it breaks down the way we want(right or left). @anuragagarwal561994 's idea !!
This issue will be dealt after #3 is completed.

create a fitness value for each trajectory ie how many blocks/ pigs it passes through.

it would be cool to have something like this research paper and an evolutionary algorithm.

Take naive agent as a base agent and create a learning table for cramming data

Detect structures on a macro level ie Tall, Wide, Dense etc. This exploits the getSupporters and isSupporter functions. We can apply a simple BFS and do this.

This should try to see how many previously played levels(including sub-levels ie after launching each bird) are similar and how much similar to the (sub-)level played currently