Deep Learning with SC2 Intro – Python AI in StarCraft II tutorial p.7

37 thoughts on “Deep Learning with SC2 Intro – Python AI in StarCraft II tutorial p.7”

1. Self.known_enemy_units includes all units as well as structures
   Self.known_enemy_structures includes only structures

2. Are the coordinates for the nexuses correct? If I compare the center of the green circle with the minimap position, it seems a bit too far down? Anyone agree?

3. Isn't it an overhead to create images first? Can't we just collect the data and feed it directly to a network?
   Also can you maybe explain a bit more about how a conv. network works and what you think could be beneficial using a conv. network over other approaches?

4. how old are youy? and how long do you know python

5. could you return to your unconventional neural nets series with sound generating or reverse engineering?

6. @sentdex
   So, first of all, those pictures you generated by yourself sure look like those from the DeepMind, I wonder why you decided to make those by hand.

   Second, coming from an SC2 player here: if you just go voidrays and optimize your strategy, you will indeed get fairly good at the game, but you will hit a ceiling. The reason is, in SC2 there is a thing called build-order, which is kind of your strategy for your early game. SC2 is a lot like rock-paper-scissors, where some openers beat the others — just because.
   You already probably noticed that your strategy kind of struggles versus early rushes. So I highly suggest to somehow program in the ability to chose a build order or something of that sort.

   There are some more things to SC2:

   First: warpgates and chronoboost. Warpgates tech is researched at the cybercore and allows you to turn your gates into warpgates, so that they ill warp and rechange faster, and also can warp anywhere in the psionic matrix. Chronoboost is an ability of the nexus, it speeds up the production of the building that you cast chronoboost on. Those two things are like really important for protoss.

   Second: scouting — the way to beat the opponent at rock-paper-scissors is to scout and predict what he will chose, you know 🙂

   Third one is called harassment, especially early game harassment — that is to deal damage with small amount of units without engaging into a battle. Create two adepts and send them right into the enemy base, best case scenario use shade (adepts ability) to get behind the mineral line and start killing workers. If you will implement build-orders I think it will learn to do the harassment on its own, otherwise I do recommend adding those two adepts at the start of the game. Alternatively you can get an oracle out of your first starport, that one also can kill most of workers with no problems.

   Last but not least, upgrades. So, there is the cybercore for air units, you can get +1 attack and later maybe +1 armor for your voidrays there, and there is forge, which lets you research upgrades for ground units. Get yourself +1 attack, it profits stalkers a lot.

7. Why were you using iterations per minute instead of tracking time by just using time.time()?

8. great video, please next video soon!!

9. Was constantly checking for a new episode lol. Thanks mate 🙂

10. where do I get a mug like that?

11. OpenAI created a bot that defeated human professionals in DotA 2 though, which is pretty complex game, does it count "solved"?!

12. Udemy is jealous right now!

13. When you said "Deep Learning" i wanted to kiss you :D.

14. when you are so smart that you need to drink water every other minute to cool down your brain.

15. Thank u for this awesome video and series

16. wow, thats how they did it in deep sc2, interesting trick

17. https://pythonprogramming.net/loading-images-python-opencv-tutorial/ Do pip install –upgrade pip and pip install C:/some-dir/cv-file.whl on the OpenCV install file (if python script dir is in your PATH var).

18. god, did you really print to know the methods in that object. I am pretty sure there's Intellisense for Python in vscode.
    https://code.visualstudio.com/docs/languages/python

19. Is it possible to train the AI bot in the SC tutorial and after that use the game easy, medium and hard bots?

20. Come for the tutorials, stay for the mugs.

21. i love your mugs

22. sucking the brains out of a cat

23. What is your editor?

24. mug game out of control

25. hey your coffee mugs are awesome

26. If you anyone needed to know about opencv2 and numpy:

    as far as i know its mostly a simple pip command and his video about how to install was outdated for me so after a bit of googling i did these things.

    pip install numpy (or pip install –upgrade numpy )

    pip search opencv ( copy paste the version with python in it )

    pip install opencv-python

    and for check in the idle of python type (or copy paste) these

    opencv =

    import cv2
    cv2._version_
    (you should see a version number here)

    numpy =

    import numpy
    numpy.version.version
    (you should see a version number here)

27. Where did you got this approach from?
    Is there a specific paper that you take as idea? Or just your own thoughts?

28. Hi, great as always. Concerning opencv: I'm on windows, using 64 bit Python. I did "pip install opencv-python" and it works like a charm. No complex installation necessary.

29. I love his coffee mugs xD

30. You and your fucking mugs make me laugh every time

31. He sure knows how to choose a mug

32. No way you actually made something sensible with this
    Wish I had the time to figure out what you actually do

33. This is gonna be fun to try out. And not to forget, laughing my ass of WTF my AI is doing 😀 Can you also do this with maps you make in the SC2 map editor?

34. the noise he makes @13.39 haha good the tutorial (Y)

35. why are my void rays just running it down mid one at a time?

36. Whatever man, let us know where you get those cool coffee mugs from! 😛

37. When you're trying to install cv2 on Mac OSX and you can't follow the series because you've been trying to install cv2 for 4 days 🙁

Comments are closed.