sighs Let us go over again why the terms are like that in DataPredict.
From Value Standpoint
-
It offers over 81 models, each have their own use cases important for games. The models are capable of impacting your live metrics. I even made tutorials related to retention and revenue.
-
Even within a model, you can apply multiple different use cases like time-to-leave and time-to-level up predictions. In contrast, plugins have very specific use cases tied to them, and at that point, it would be acceptable to be for it one-off.
-
You now no longer need PhD-level knowledge to implement a whole full-set of ML. I have lowered knowledge barrier for you that could take up years to understand the whole math behind it. Not to mention, knowing math != knowing how to implement it. Two separate skillsets. Even from OpenML, I already see issues with his deep q-learning, but he isn’t here right now to respond. These are not simple bugs but rather issues of understanding the maths behind it, and so it is critical to get them fixed. You can lose credibility quite quickly if your implementations aren’t correct, especially if you use them as part of live system.
-
If I were to ask you to implement to something, let say… deep reinforcement learning for self-learning AIs, you would take a month to get this right (which can be shorter if you have Masters or PhD background). Not to mention, you have no alternative for smarter AIs except for more rule-based AI.
-
Now compare this to a generic plugin where it has one use case like MoonAnimator, where you can just live without animation or do some work around if you never purchase the plugin.
-
-
You now no longer need a headache to learn another programming language and then learn a bunch of many different machine learning, deep learning and reinforcement learning frameworks. I have unified it into one and only choose features that are usable and have valuable use cases for games.
-
I have repeated this almost every single time: Free to use for companies making up to 3K USD revenue within a year. Plus, the 2% decreases as you earn more revenue.
From Cost Standpoint
-
3 years worth of development, which would be equal to 300K USD in time cost. This also includes trying to make sure it matches the academic standard. Unlike your usual game programming, bugs can be hidden in numerical values even if your mathematical implementation is correct. So, I spend much more time debugging more for each features just to make sure academics and commercial users does not scream at me that “it does not work”. I even went far asking other researchers who are more well-versed in these subjects to make sure my implementations are correct.
-
Machine Learning Engineers (or rather ML Framework Engineers) for Lua are long gone since Torch7 library is dead (Torch7 is what gave birth to PyTorch). I am the only one who develops the whole suit of models in Lua (I’m not excluding other potential Lua ML projects, but I talk in terms of academic-grade and production-grade standard).
-
If I were to say that I have ML engineers who could work alongside with me, it wouldn’t have such terms. And I have been working on it for three years solo. I doubt you want to do the whole library yourself from scratch. And I made sure each aspect of the library have their own use cases. I didn’t add certain features since it did not add any value to game-related use cases.
-
You can’t just pick a random Machine Learning Engineer and expect them to know how to implement the whole thing in Lua from scratch. Their jobs are to implement machine learning models from existing frameworks to their projects. Emphasis on “existing”.
-
Additionally, most ML frameworks in Lua solely focused on deep learning, not general machine learning. So, I’m really question anyone who has all machine learning, deep learning and reinforcement learning knowledge in one place for Roblox. There’s no scikit-learn for Lua before this library, after all.
-