Feedback and confusion

[The Crimson Fox]

I've played the game for two hours so far, have gotten through several of the levels, but I'm left confused over a few things.

(1) The game says it's about machine learning, but all it has me doing is hooking up lines to boxes. "Program this car to recognize red lights!" Sounds great! ... but all it is, is hooking up a "red color" line to the output and the "green color" line to a trash can. Machine learning involves more than that ... right?

(2) What am I supposed to do when some of the outputs overlap? Like, put red/blue/green into this output, but put only blue/green into this other output. The color test only lets me specify two colors; the third color will leak into either output, unless I add additional tests which make it take too long and not complete. Is it acceptable to let bad inputs through and have a lower accuracy score in the output?

(3) Are "correct" solutions posted anywhere? I don't know whether I'm approaching the challenges the right way, and the game doesn't really offer much guidance.

[Elonmusk]

seriously im right there with you. i play for a few minutes. im a programmer and i was hoping this was more than it really is. sadly i don't think this has anything to do with machine learning i think its just a puzzle game.

[The Crimson Fox]

skyprog, after putting another hour into the game. I agree with you. I'm also a programmer, and I think this is really just a puzzle game.

It *could* be so much more. Instead of shapes, it could use "images captured by the automobile's LIDAR"; and instead of simple color tests, it could offer a variety of machine-learning algorithms which have various strengths at recognizing images. That way at least the puzzle would actually have something to do with its description, and it could at least seem more like a real-world solution - get people actually familiar with the kinds of machine learning tools that are used in the real world.

And I'm getting to the point where I'm confused by some parts of the game. What are DLLs? What are CatHub repos? I get one screen of information immediately before I have access to them, and then nothing. I've looked all over for any sort of explanation of what they are. I can't figure out if CatHub repos are just where I can store different attempts at solving the problem, or whether they are all run in parallel so i should copy my best solution to all of them.

Some of the puzzles are just a bit too slow and I don't see any way to get golds on them. I don't know if this is because I haven't found the right answer yet, or because later I'll unlock a powerup that makes the puzzle easier.

[ElJorro]

Hey there fellas!

The Crimson Fox, thanks for your very detailed feedback!

Let me try and answer those questions from your initial post at first

1 - It is a puzzle game, no secret there. In fact, it's written right in the description :) However, it WAS developed by people with a machine learning degree (not including yours truly, unfortunately) and they were actually trying to make the experience as close to real life as possible. We were and still are dreaming of making an honest machine learning game, but that would mean listing an actual Google server in system requirements... So, we've decided to go for an approximation of machine learning, because our goal was to help people understand the general concept of this field of knowledge (not all of our players are programmers, you know ;)) By the way, those set of levels about a self-driving car actually DO have machine learning in them, although a bit simplified

2 - Yes, a certain percentage of errors is acceptable as long as you scheme reaches the target percentage of correct data

3 - Yes! For starters, try our Discord server: https://discord.gg/W8KhjSf

[ElJorro]

DLL's are your schemes from your previous tasks that you can use as one node to make your life a bit easier and not having your work field clogged with all those nodes. Not sure if that's a perfect analogy, but you can think of them as shortcuts to some schemes that you've created in the past

CatHub allow you to have more than one solution to a problem at the same time and experiment with them as you please :)

I hope that answers your questions, but let me know if you have more! Small Cat is watching you ;)

[The Crimson Fox]

Thank you very much for your reply, Developer! :-)

The game's description says "learn how machine learning really works" - I'd say this is as accurate as saying that Civilization teaches you politics. It doesn't. Civ only lets you use the end result of politics, and WTL only lets you use the end result of machine learning. Nothing necessarily wrong with that (Civ is an epic game!).

But I think WTL would be a much more interesting game if (1) the puzzles were closer to real-life situations instead of sorting shapes and colors every time, and (2) more attention were given to the actual machine-learning boxes. What are the challenges in detecting the difference between squares and circles, and green and blue? Why does a machine-learning algorithm need to be trained, and what does the training process actually entail? How does ML decide what parameters are important in its decisions? (Don't answer me here, just keep these in mind for future revisions.) I think it's great that the game has links to articles and YouTube videos about various ML topics, but it's difficult to understand those big dry topics in context of the game.

(Take Kerbal Space Program as an example. It gives you challenges to solve and rocket pieces to solve them with, but it feels very much based in real-life spaceflight, with explanations of the physics problems involved and the consequences of various choices. After playing a few missions in KSP, I feel like I have a better understanding of the space program, and like I can watch a real-life rocket mission and know what's going on. With WTL, I don't feel like I could actually apply machine learning to a real-life problem yet.)

Now that I see WTL as a puzzle game, I think I can enjoy it more, while at the same time being disappointed for the machine-learning exercise it could have been. But I think that's still within your grasp, and I hope you eventually make it all the way there!

Oh, and regarding your other answers:

- Thank you for the explanation of DLLs and CatHub! Putting that help information into the game would be useful - like, a help button in the DLL screen and next to the CatHub tabs. Right now, after the first time I see the information, it's gone (as far as I could tell).

- And it might also be helpful to provide solutions to some common situations. Perhaps, after the player has solved a few puzzles, unlock a tips screen saying 'if you run across these situations, here are some efficient solutions to them that you can use.'

Thank you for your replies above, and thank you for the game! I wouldn't have written so much if I didn't enjoy it as much. :-)

[GospodinNoob]

Thanks for the great feedback!

[Dedmoin]

I also understood the game is about get to know ML techniques and go into the topic into more depth, based on the description in Steam store.
Because this would be just amazing. I already own a couple of programming puzzle games that managed to be challenging yet using simple instructions while introducing players to (simple) programming techniques and approaches.
I know ML uses more complex mathematical calculations but those could probably be explained and created as 'DLL' and abstracted this way. Also in real live I ought to know how to calculates matrixes but avoid to do it manually :D

nice game ;)