How to use tensorflow in godot with gdscript

vmjcv

I designed a small game with novel rules. I want to add AI to the game. If I write FSM to achieve it, it is a very huge workload, and I will add a lot of new games with novel rules. I implemented part of the code with reference to AlphoZero , but I encountered difficulties in the logic of interacting with tensorflow. I hope that tensorflow can be used in godot,

I refer to the following: https://github.com/godotengine/godot/issues/30613 https://github.com/godot-extended-libraries/godot-tensorflow-workspace

But they seem to need to change the C ++ code and recompile every time they modify the model. I hope that I can use gdscript to write the model directly without recompiling like python.

Are there any suggestions or solutions that can be provided to me

i need your help please,thanks a lot. BR.

SIsilicon28

Why would you want to use Tensor flow instead of a FSM? How would writing an FSM create a "huge workload"?

TwistedTwigleg

I agree with SIsilicon28, I'm not sure I see the benefits of using Tensorflow over a Finite State Machine (FSM).

Regardless, since Tensorflow is a C++ library, you'll have to modify the C++ code to use it within Godot, or use a binding like the Godot-Tensorflow-Workspace you linked to. It could be possible to bind Tensorflow to Godot through GDNative, though I have no idea how feasible that would be.

You could also try to use something like TensorFlow.Net with the Mono version of Godot. I'm not sure if it would work though, especially depending on the dependencies of Tensorflow.Net You could also, potentially, use Godot-Python and Google's provided Tensorflow API

However, from a performance point of view, using C++ would probably be ideal. But there might be some other options to use Tensorflow without C++ that could be worth investigating.

vmjcv

I briefly explain the rules of the game, although it is boring.

I have a game idea that combines tic-tac-toe and cube, Each round, the player can add squares or rotate the cube to the block, and can only do one operation. When a player first completes the three-child continuation of a certain face, he wins.

The special rules are: 1. Do not do the reverse operation of the last player 2. Do not add a grid to the grid that already has a block 3. When a certain position is for two players to win at the same time, the winner is not the last player to make the operation 4. Player1's block style is xxo, xo, x .Player2's block style is oox, ox, o

The operation to add a grid is as attach(I cannot add image here, so I generated a zip package)

I think if using fsm to implement ai is a very heavy workload, and I don't quite know what the winning strategy is.So I want to use code similar to AlphoZero @SIsilicon28 @TwistedTwigleg

vmjcv

First，

It could be possible to bind Tensorflow to Godot through GDNative, though I have no idea how feasible that would be. I really want to try to bind Tensorflow to Godot via GDNative, because I want the main code in gdscript.

Second，

You could also, potentially, use Godot-Python and Google's provided Tensorflow API I am a greedy person, I want to know if Godot-Python can use both python and gdscript in godot

Thank you very much for your answers.@SIsilicon28 @TwistedTwigleg