| Sanqui 2060 ![]() ![]() ![]() ![]() ππ€ππ€π¦ Level: 87 Posts: 1892/2066 EXP: 6305066 For next: 87708 Since: 12-20-09 Pronouns: any β¨ From: Czechia | Estonia Since last post: 6 days Last activity: 1 day |
|
||
|
*Sanqui flops, wings involved. level 82 exp 5201630 exppct 96.5% numposts 1892 ![]() Sanqui's back at it again! Today I'm learning to use LSTM cells, or "long short term memory" cells, in order to work with temporal data. In particular, I want to teach a neural network to predict the next frame of a Super Mario Land gameplay. And maybe another, and another... and see what it comes up with? Why SML and not SMB? The Game Boy screen resolution is smaller and there are only four shades, which means simpler model and faster training, which means better results and I can learn and apply these techniques to more complex games later. I'm really curious to see if the neural network can learn some physics or level geometry. I've previously made a video with Sonic 1, if you haven't seen that yet. However, this time around, don't want to work on merely pixel data, I want to experiment with tiles and sprites as well. As well as input data, so that maybe we could predict the next frame given the input... So, I'll need YOUR perfectly average gameplay movies to teach the AI! If you wanna help, download my readymade .zip and follow these instructions: 1. Unzip the package somewhere 2. Launch EmuHawk.exe 3. File > Open ROM 4. Locate and choose Super Mario Land (JUE) (V1.1) [!].gb 5. Play a bit to familiarize yourself with the controls: arrow keys=d-pad, Z=B (run), X=A (jump) if you need it, use ESC for pause (NOT enter/start), use backtick (`) for rewind 6. Movie > Record Movie, OK 7. Play the game upon completion. If you want to, you can rewind, but do leave some deaths in, it should look and feel like a normal playthrough. If you get bored playing, I'll accept a partial playthrough too. 8. File > Movie > Stop movie 9. Close the emulator 10. Send me the file BizHawk-2.5.2/Movies/Super Mario Land (World) (Rev A).bk2 It shouldn't take you longer than half an hour to beat SML. I've already got contributions from three people, I'll need a few more though. Thank you for helping me help my AI help us all! ![]() ____________________
In our brief lives, β’ we've managed to meet. β’ Treasure this gift, β’ this precious time that we have.
|















