[Added SAC, DDPG and TD3] DataPredict [Release 2.0] - Machine Learning And Deep Learning Library (Learning AIs, Generative AIs, and more!)

Resources Community Resources
, ,
#745

Release Version 1.18

Added

  • Added ValueScheduler class. These will help you adjust the values as you call the calculate() function.

  • Added setEpsilonValueScheduler() and getEpsilonValueScheduler() function into the ReinforcementLearningQuickSetup.

  • Added setLearningRateValueScheduler() and getLearningRateValueScheduler() function into BaseOptimizer.

Changes

  • Renamed setPrintReinforcementOutput() to setPrintOutput() for ReinforcementLearningQuickSetup.

Removed

  • Removed epsilon decay factor parameter inside the ReinforcementLearningQuickSetup in favour of using ValueScheduler.

  • Removed timeStepToDecay parameter from the LearningRateTimeDecay optimizer.

1 Like
#750

Hi. Does this allow reward learning?

#751

Somewhat. Currently the library only have Random Network Distillation if you want to do internal rewards.

#752

Im trying to create something like in those videos of how it has to walk if yk what i mean

#753

Possible, but quite limited with this library. Currently, it doesn’t support continuous action spaces. Only discrete ones.

Likely it won’t be implemented on this library, but rather on “DataPredict Neural” library.

#756

Can I make a personalized advertisment ai that recommends new products in my shop with discounts based on previous user purchases, chat history, and product viewing behavior

#757

Absolutely can. There’s a plenty of algorithms you can choose from.

  • Clustering Models? Yes.

  • Neural Networks? Yes.

  • Neural Networks with Reinforcement Learning? Hell Yes.

1 Like
#758

which algorithm would you recommend for maximizing the purchases? I’m planning on potentially using another ai too in picking discounts.

For example discounts are given to increase spending. So the ai might learn specific patterns of discounts to maximize spending.

If a user never spends then they get big discounts. IF they do spend but require certain patterns, the ai learns.

1 Like
#759

Maximizing Purchase? Probably stick with reinforcement learning. Those models will try to maximize getting rewards (in this case the amount of purchase). Though it might take a while to train.

As for discounts, be careful. You might not want the user exploit the model just by selecting discounted items, but never undiscounted items.

1 Like
#760

I might use ai for personalizing objects that maximize purchases and a simple algorithm I script myself for selecting discounts. The discount will be based on their previous purchases.

Never spent: 80-90% off
Rarely spends: 10-30%
Occasionally spends 5-20%
Consistent spending 15-40%

Frequency of previous purchases is accounted based on a point system, with older purchases degrading. Once a user makes their first purchase, they can never return to the “never spent category.”

1 Like
#766

This post didn’t age very well…

3 Likes
#779

Heads up guys!

The next update will allow some of the algorithms to support continuous action spaces!

I’ll be updating the Beta version multiple times before releasing a stable release version for all to use!

So get ready!

Why I do this? Well, my plan is to make DataPredict as the industrial and research standard for RL in Roblox. So I’ll be completing this update before I leave for my Masters.

3 Likes
#780

Release Version 1.19

Added

  • Added DiagonalGaussianPolicy and placed it under QuickSetups section.

  • Added a new parameter for reinforce() function to AsynchronousAdvantageActorCritic model.

  • Added diagonalGaussianUpdate() function to AsynchronousAdvantageActorCritic model.

Changes

  • Renamed ReinforcementLearningQuickSetup to CategoricalPolicy and placed it under QuickSetups section. Also made some internal code changes.

  • ReinforcementLearningBaseModel’s and ReinforcementLearningActorCriticBaseModel’s setUpdateFunction() and update() functions have been replaced with setCategoricalUpdateFunction(), setDiagonalGaussianUpdateFunction(), categoricalUpdate() and diagonalGaussianUpdate().

  • Made internal code changes to all reinforcement learning algorithms in the library.

  • Made a few API breaking changes related to the AsynchronousAdvantageActorCritic model:

    • Renamed update() function to categoricalUpdate().

    • Renamed reset() function to resetAll().

    • Renamed singleReset() function to reset().

Side Notes:

Please update the MatrixL library so that you don’t run into issues when using this DataPredict library version. Some changes have been made at MatrixL library and these changes gets transferred over to the DataPredict library.

2 Likes
#781

I also have added a new tutorial to the documentation that explains the discrete and continuous action spaces. Go on and have a look!

4 Likes
#788

Can u share it for us? I havent taken statistics class

#789

Can you profit free tools that implement ur module? Stuff like anti cheat, preference selector (from a table of items, what they like), etc? Ive never taken statistics class and idk ai stuff so i cant use ur module. That or u could give general tutorials on how ai works or how statistics works

#790
  1. Additional questions:

  • Will that free tool be distributed publicly? Or is it for internal use?

  • You already have ChaWatcher, which is partially open source when it comes to anti-cheat. You can only use that algorithm inside ChaWatcher for cheat detection only. Why not use that?

  1. No. You’re basically asking me to teach you for free. And the whole AI thing can’t be covered in few posts, but in a year. Even six month worth of university lessons doesn’t cover the whole library.
1 Like
#791

I wish Roblox would support GPU acceleration in the future lol to support your work.

#792

At least we have coroutines. We can still speed up the training time by distributed training to replace the lack of GPU acceleration.

1 Like
#794

You can also utilize the inferior version of GPU cores, the CPU cores. By using actors to parallelize intense computations by spreading the load into many actors, you can do computations in parallel and fetch them for use. This proved effective when I tried to make my ant simulation faster

2 Likes