DeepGamingAI_FIFA icon indicating copy to clipboard operation
DeepGamingAI_FIFA copied to clipboard

Players positioning

Open giovannicaprio opened this issue 6 years ago • 2 comments

Hello,

Very nice work. I have some questions.

  1. How you figured out where a player is in game? I mean, if he is in attacking position or defending position.

  2. Would you able to keep a track of players storing a position of each of them in each frame and knowing exactly who was the player?

How did you differenced teams?

giovannicaprio avatar Jun 06 '18 18:06 giovannicaprio

Thanks for the question. I've taken a Machine Learning approach here instead of the traditional AI approach, which means instead of hand-crafting rules of the game, I let the Deep Learning model learn them by itself.

For simplicity, try to compare this with the image recognition problem. Earlier, we used to manually extract features like HOG from images and then make classification model using these. However, with Deep Learning, we do not need to manually extract these features and we instead let the model learn them over time, hence they need a lot of training data to perform well. Similarly for this AI bot, we are not doing any feature engineering on player positions, game rules, etc. but letting the model learn all that from training data.

You may find more about why I chose this approach here: https://towardsdatascience.com/building-a-deep-neural-network-to-play-fifa-18-dce54d45e675

Hope that helps, cheers.

ChintanTrivedi avatar Jun 11 '18 00:06 ChintanTrivedi

Thanks for answering @ChintanTrivedi . I understood your approach, I think you are correct on doing that. I had already read the post you mentioned, it is very good. I asked about players positions be cause I am interested in extract informations from that positions, for example. Make a graph, or something. But as far as I understand, you do not have control of this information. Correct?

giovannicaprio avatar Jun 13 '18 20:06 giovannicaprio