spec.md

create an environment where agent is given a pg essay text and it comes up with a compression and decompression algorithms for it. the environment runs the algorithm on essay and gives +1 reward to agent if curent step's compressed text size is lower than avg of all compression sizes achieved so far in previous steps for that essay. it also runs compressions and decompression on the essay to verify that compression and decompression algorithms work correctly, if they don't then it's -1 reward. it also runs state of the art eg zip, bzip etc top text compressions on the essay and checks the size. if agent achieves smaller size than any of them it's +10 and if it achieves smallest size than it's +20 reward.