Post a link to a playable game, preferably WebGL. If you don’t have a playable game, post something substantial, not just text.
How To Ask For Feedback
Be concise.
Specify what you want feedback on and what you don’t.
Resist the urge to write an immediate defense. Take the time to understand their points. Remember that your friends here are taking time out of their busy schedules to help you for free.
How To Give Feedback
Be positive. There’s something of value in every game.
You can leave those details empty, there’s a PNTD (Prefer not to disclose) option. The reason I collect that is because gender and other details (on average) may have an influence on what is found to be interesting to players.
You’re asked to rate each level with a score from 1 to 6. where 1 is terrible and 6 is great. Th reason for a 6-star rating is you can’t be exactly neutral, 3 is a bit negative and 4 is a bit positive. You can also optionally leave some remarks on a level.
So, what this build is, is a uniform selection of levels made via k-means (k=30) clustering. The clusters are made on a wide variety of metrics on the level (Average number of neighbouring platforms, visual linearity, calculated (A*) length to the goal, number of mechanics, etc.) The purpose of the project is to establish what is interesting to players based on those metrics.
After this experiment I use the metrics and player data to train a ML model (Neural Network (Deep Learning), SVM, Decision Trees, etc.) to find correlations between what is interesting to players and if I can optimise to make it more appealing to a larger target audience (every gender, every age group, etc.). Or I could find that the interests of people are too different and optimise for a smaller audience.
But I don’t have enough data yet to tell if there’s anything significant.
I’ll be publishing the findings in my master thesis.
Interesting concept, but I’m not sure I understand how you constructed the hypothesis that these metrics would vary with the variables you described?
I would imagine that something like the Big 5 personality traits would be marginally more relevant to such metrics as the distance-to-goal and number-of-mechanics than education level. Although I still would imagine that’s too much of a stretch to get useful data from, and would suggest a bigger variety of clearly defined mechanics of the sort that really tend to polarize people one way or the other, such as aggressive/defensive, logical/intuitive, twitch-based/reasoning, sharp/shallow difficulty curves etc.
An interesting thing about big data and number crunching is that you can just gather interesting data points and see what patterns emerge without necessarily devising a lot of hypotheses ahead of time. Of course the data points you choose to collect can also produce unintended biases in results, which does make people a bit wary about looking at things like gender.
@coen22 - As for the gameplay, the camera drove me a little crazy. The player should either have control of the camera, or it shouldn’t rotate on its own separately from player control during play. Addressing the camera would probably increase your response rate dramatically. You could either make it always follow behind the player or use a fixed orientation.
That’s a fair point, the camera is now controlled by the AI to send you in the right direction, but it’s behaviour is not quite as natural as I intended. I was told to leave it in, because beforehand it just followed in a Mario 64. like manor (which I was told wasn’t nice). But of course people from uni aren’t gamers.
Yes it’s probably something I’d do different the next time round. But I was running a bit out of time. Metrics on the levels are important though, to understand (and enforce) if the variety between levels is actually decent.
Not knowing how AI generates level and what inputs is getting, is hard to look at the project objectively.
From pure game perspective, is bad as its gets.
Moving camera is the first issue, as already appointed.
There is input lag.
Lack of shadow, makes player disconnected from the world. Unable to tell intuitively, where player is in 3D space. Specially when during jump.
I haven’t finished first level. Gameplay is too random to me (not talking about islands generator).
I am not getting, why not let camera being controlled by mouse input. Or by any other manual means?
I am not sure why AI is needed here? How AI select the direction? How it is different, from getting nearest island?
Same result can be achieved using path finding algorithms, by taking nearest node.
What is special about AI generated levels in your case?
You could equally use some noise (i.e. Perlin noise) and other randomizer, to achieve very similar results.
As for master thesis, I haven’t seen paper of course, but is very lacking in my opinion. Project is approached very vaguely.
