Arbitrary Error?

[Mitch Rocks]

I keep finding scenarios where I can get better than needed accuracy, but the time limit doesn't let enough nodes through for enough correct nodes to fill unless I use a single node and send every last packet straight through.

I went into this expecting the accuracy to be a minimum, not a maximum, and it feels unintuitive to have to make my program worse at what it does so it can hit enough false positives to fil the bar.

It feels like being punished for trying to make the code better, and the unintuitiveness leads to me struggling to figure out which foot I should shoot myself in to make my program work.

For example, Task 26, RC75%, has a 75% error "margin" and 65 seconds, during which 131 node are spat out at random. It takes 39.2 seconds for a decision tree color, fastest program I have at 0.286 s, taking all the input from source with both outputs on the program's output to send all nodes from start to end.

If one node which is the fastest one I have has only 25.8 seconds to spare, then actually sorting through the data to hit the accuracy requires strategically sabotaging myself, making my code worse and intorducing inaccuracies for the sake of a strict constraint on nodes processable. It also doesn't help that I can't parallel process for the task in question.

Task 25 is more akin to what I expected. I had to buy decision tree, but I could use it to make Only Square 60% perfectly accurate.

Like, I get that not only optimal solutions can be right IRL, but if that's the intent, A: Perhaps have it be stated in text instead of only by the mechanics (as percents have been heavily hammered into me as a minimum, not as a maximum, and no doubt the same is true for others), and B: perhaps instead of making a good optimal solution be mathematically impossible, perhaps have it just be highly complex, with a more obvious and simplisctic non-optimal means of solving. Sends the same message, but doesn't frustrate a player by making the sub-optimal solution extremely complex while having a simple optimal solution restricted by unreasonable parameters.

Failing that, perhaps make it so a player can forfeit a Task, forfeiting the money it would create as well. If the task would create a node required in a later task, have it added into the store to purchase with in-game money.

[GospodinNoob]

But, this margin is a minimum! Not a maximum, if you reach 100% accuracy with 75% requirements it's ok! It can work slower, it's true. The idea in this task looks like: minimize nodes number and reach minimum accuracy, not: reach 100% and corrupt the scheme.

In future we will add some tutorials about that, thanks)

[Mitch Rocks]

Originally posted by GamineBrownies:

But, this margin is a minimum! Not a maximum, if you reach 100% accuracy with 75% requirements it's ok! It can work slower, it's true. The idea in this task looks like: minimize nodes number and reach minimum accuracy, not: reach 100% and corrupt the scheme.

In future we will add some tutorials about that, thanks)

My point is that tasks like 26 make it mathematically impossibe to actually do so.

There is no "if you reach", and that's where it feels wrong. It feels less like a lesson and more like playing a game that cheats.

If it feels like the sample packets are stacked against you from the start, it's quite demotivating.

In task 26, there's no guarantee that 56, or even just 50, correct nodes will be sent, meaning that you need to settle for 42 (75%) and have 14 false positives, or you'll probably autofail, unless you get super lucky when releasing.

[Mitch Rocks]

There's also task 16, which shows it more easily. 40 green squares, 41 needed for 100% accuracy rate. Forcing players to be suboptimal in the accuracy give a feeling of hurting good performance. Accuracy requirements being less than 100% seems at first like a safety net, but I've reached a point where I dread having to do less than perfect, because for some reason, perfect is easier than not.

If the sample packets don't even contain enough right packets for an instant sorter to pass the test, it feels like the parameters are incorrect.

Task 26 is just a matter of the player needing a moment longer to realize that there is no going perfect here, since even though insta-sort could pass it, as there's technically enough good packeta to do so, there are so many bad packets that the stream is too polluted for anything except sheer luck to get accuracy perfect while still not running out of time with the tools that the player has.

Tldr, before even starting the game players expect the accuracy percentage requirement to be difficulty level, I.E. the lower the percent, the sloppier they can get away with. That appears to not be the case, and the game doesn't do well with getting players used to going against the thought.

[GospodinNoob]

In real life machine learning task sounds not like "we need 100% accuracy, or you will die", it more like "we need more accuracy than we have without machine learning, 3% advantage will be good". S we trying to explain, that solution wih accuracy level lower then 100% is ok) Thank for your feedback, we will discuss this problem in our team. Really, i understand your point of view, and i will try to change game logic and tasks to way which erase this misuderstanding.

[Mitch Rocks]

Originally posted by GamineBrownies:

In real life machine learning task sounds not like "we need 100% accuracy, or you will die", it more like "we need more accuracy than we have without machine learning, 3% advantage will be good". S we trying to explain, that solution wih accuracy level lower then 100% is ok) Thank for your feedback, we will discuss this problem in our team. Really, i understand your point of view, and i will try to change game logic and tasks to way which erase this misuderstanding.

Thank you. It feela like the idea of " less than 100% is acceptable" would be a point best implied through mechanics or stated in game text, rather than made as neccessary with no text explanation, unless this game is intended as a training simulation for actual machine learning developers, but all in all, pretty good game! Looking forward to more updates!