Faster learning and vizualization

[ZeeWanderer]

• Add argument parsing to flappy.py
  • --fps allows to set FPS for faster or slower vizualization. Since Framerate is not used in computing player position the results are correct.
  • --dump_hitmasks dumps hitmasks to file end exits. Hitmasks are dumped from random pipe and player but since all sprites of the same type differ only by color it doesen't matter. This is required for learn.py
  • add dumped hitmasks file hitmasks_data.pkl since it is correct for all cases unless default sprites are changed.
• Add optional arguments to update_scores and dump_qvalues in bot.py Needed for learn.py
  • dump_qvalues in update_scores allows not to dump q values.
  • allways_dump in dump_qvalues allows to allways dump qvalues
• Add learn.py Basically this is a lightweight script purely for learning. It lacks any visualization and works a lot faster then flappy.py.
  • Usage: python learn.py [--cmd_out] <number of iterations>.
  • --cmd_out option outputs (iteration | score) to cmd after each iteration.
  • All image parameters like IMAGE['pipe'][0].det_height() are substituted by their actual values from default sprites.
  • hitmasks_data.pkl is needed because .convert_alpha() cannot be called without initing pygame.display so the apha channel is not blended which results in incorrect hitmasks.

Illustration


[chncyhn]

Thanks @maximilian578 these look great! I will look into the PRs tonight. 👍

[ZeeWanderer]

@chncyhn, by the way why qvalues are dumped each DUMPING_N Iterations? The qvalues.json is truncated and written anew each time qvalues are dumped. It would make sense to utilize pygame QUIT event and dump qvalues only when game is exited, but then closing interpreter by closing editor or console will result in qvalues not being saved. atexit.register() would probably not work either when closing interpreter.

[chncyhn]

@maximilian578 honestly it's a matter of convenience. I have forked the Flappy Bird clone from sourabhv/FlapPyBird, and am not aware of pygame's internals. Dumping q-values at every N iteration is just a very simple solution which works. For this project's purposes, the simpler the better.

[chncyhn]

@maximilian578 really cool stuff! I've just changed a couple of things in naming. Merging! 🎉