RingML: High-Performance Machine Learning Library for Ring

Azzeddine Remmal

unread,

Dec 10, 2025, 12:19:00 PM (9 days ago) Dec 10

to The Ring Programming Language

Hello Mahmoud

To build a modular, high-performance Deep Learning library in Ring language, leveraging the FastPro extension for accelerated matrix operations. The goal is to provide a syntax similar to PyTorch/Keras for building Neural Networks and eventually Language Models.

RingML: Deep Learning Library for Ring Language

I've encountered some issues with FastPro; I hope we can fix them in the future.

RingML Project & FastPro Technical Analysis Report

Greetings, Azzeddine

AgentCodeGear_EN.md

Mahmoud Fayed

unread,

Dec 10, 2025, 6:32:17 PM (8 days ago) Dec 10

to The Ring Programming Language

Hello Azzeddine

Thank you very much for starting this project :D

With respect to RingFastPro, and the related functions and the issues that you want to fix

Feel free to update the library source code, and send pull requests: ring/extensions/ringfastpro at master · ring-lang/ring

Just keep the current API & samples working without changes

i.e. Test changes against the next samples: ring/samples/UsingFastPro at master · ring-lang/ring

Many of recent FastPro features are influenced by your neural network sample: ring/samples/General/NeuralNetwork at master · ring-lang/ring

So, consider this part of the library like your code and help me and Bert to improve this library

Also, beside testing your changes against the sample, you could test the next apps by Bert and Youssef, to be sure that your changes/fixes doesn't cause problems

ring/applications/imagepixel at master · ring-lang/ring

ysdragon/im2ansi: Convert images into beautiful ANSI/ASCII art

Keep up the GREAT WORK :D

Greetings,

Mahmoud

Mansour Ayouni

unread,

Dec 10, 2025, 6:39:45 PM (8 days ago) Dec 10

to Mahmoud Fayed, The Ring Programming Language

Hello Azzeddine,

I also use FastPro extension in Softanza in this file:

https://github.com/mayouni/stzlib/blob/main/libraries/stzlib/base/number/stzFastPro.ring

In fact, I made a semantic reengineering of the available operations to fellow a unified pattern:

https://github.com/mayouni/stzlib/blob/main/libraries/stzlib/base/doc/narrations/stz-fastporupdate-narration.md

Therefore, if you change the extension structurally, I need to update my code.

Keep working on this nice RingML staff, we all need it.

All the best,

Mansour

--

---
You received this message because you are subscribed to the Google Groups "The Ring Programming Language" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ring-lang+...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/ring-lang/7ce05105-c251-4a99-8398-700b2370fde8n%40googlegroups.com.

Mahmoud Fayed

unread,

Dec 11, 2025, 2:02:04 AM (8 days ago) Dec 11

to The Ring Programming Language

Hello

RingFastPro package is updated to include recent changes by our friend Azzeddine

To get the update (So, you can test it with your code)

ringpm update ringfastpro

Greetings,

Mahmoud

On Wednesday, December 10, 2025 at 8:19:00 PM UTC+3 Azzeddine Remmal wrote:

Mahmoud Fayed

unread,

Dec 11, 2025, 2:23:10 AM (8 days ago) Dec 11

to The Ring Programming Language

Hello Azzeddine

In Ring applications, we have Chess End Game: ring/applications/chessendgame at master · ring-lang/ring

Description: https://arxiv.org/ftp/arxiv/papers/2111/2111.05976.pdf

You will find the dataset in this file: ring/applications/chessendgame/data/chess.csv at master · ring-lang/ring

The goal is to predict the game result (draw, win in one move, win in two moves, ..., win in sixteen moves)

i.e. it's a multi-classification problem

And we have a GUI app that display the chess board for each case in the dataset: shot1.png (1328×645)

Suggestion: In RingML project, you could add an example about predicting the game result and add it here: RingML/examples at main · azzeddine2017/RingML

It will be an interesting example, because it covers multiple tasks

1- Reading the CSV file (using CSVLib.ring)

2- How you will process the data-set (dealing with text inside it, not just numbers)

3- Multi-classification problem

You could do this as a console app (like other samples) where we have a program where the user enter the data

* White King File

* White King Rank

* White Rock File

* White Rock Rank

* Black King File

* Black King Rank

Then the program provides the predicted output (Game Result:) for this case using the Model ---> Draw, win in one move, win in two moves, ..., OR win in sixteen moves)

Greetings,

Mahmoud

On Wednesday, December 10, 2025 at 8:19:00 PM UTC+3 Azzeddine Remmal wrote:

Bert Mariani

unread,

Dec 11, 2025, 8:57:29 AM (8 days ago) Dec 11

to The Ring Programming Language

Hello Azzeddine

Thanks for reporting the issues and providing the updates to RingFastPro !!

Using it for your ML Machine Learning project will be of great interest.

I will go through the Test Scripts and other Apps that use Ring Pro to verify.

Best Regards

Bert Mariani

unread,

Dec 11, 2025, 9:37:09 AM (8 days ago) Dec 11

to The Ring Programming Language

Hello Azzeddine, Mahmoud

C:\ring\applications\imagepixel\ImagePixel.ring # Works

C:\ring\samples\UsingFastPro\MandelbrotAnimate\MandelbrotAnimate.ring # Animation does Not Work

It will draw the Initial Mandelbrot image

User Selects the Smaller Square to zoom into

Does Not Do the Amination part of zooming it

Just Displays the the Selected Area - static.

Mahmoud Fayed

unread,

Dec 11, 2025, 6:17:13 PM (7 days ago) Dec 11

to The Ring Programming Language

Hello Bert

>> "C:\ring\samples\UsingFastPro\MandelbrotAnimate\MandelbrotAnimate.ring # Animation does Not Work"

These are the updated done by Azzeddine: Update matrix operations to use nValue by azzeddine2017 · Pull Request #1594 · ring-lang/ring

Could you check them and see which part have an effect on MandelbrotAnimate?

Greetings,

Mahmoud

Mahmoud Fayed

unread,

Dec 11, 2025, 11:26:33 PM (7 days ago) Dec 11

to The Ring Programming Language

Hello Bert, Azzeddine

>> " MandelbrotAnimate.ring # Animation does Not Work"

I have already tested the sample, after changing the current option in the combobox from (Draw Point) to any other option like (Mariani) then clicking (Draw), I see the animation works :D

Greetings,

Mahmoud

On Thursday, December 11, 2025 at 5:37:09 PM UTC+3 Bert Mariani wrote:

Azzeddine Remmal

unread,

Dec 12, 2025, 8:07:22 AM (7 days ago) Dec 12

to The Ring Programming Language

Hello Bert and Mahmoud

Thank you for checking the test scripts and other applications.
I was busy implementing Mahmoud's suggestion to add an example about predicting the end result of a chess game.
I'm working on a low-spec machine; training the model takes a long time, and verifying everything is crucial.
I tried using Google Colab, but it didn't work.
The compiled version of the language is running on a recent version of Ubuntu,

and Colab is running an older version.

Even building from source didn't work; there were many errors.
In the end, I had to reduce the model's dimensions and complete the training on my machine.

The great news is that I was able to prove the principle of operation, and I will share the results as soon as they are complete.

Regards, Azzeddine

The Future of Programming

unread,

Dec 12, 2025, 8:12:34 AM (7 days ago) Dec 12

to The Ring Programming Language

Hello Azzeddine

You are welcome :D

Greetings,

Mahmoud

--

---
You received this message because you are subscribed to the Google Groups "The Ring Programming Language" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ring-lang+...@googlegroups.com.

To view this discussion visit https://groups.google.com/d/msgid/ring-lang/67922284-1b31-40ae-8f6b-012d2d446edan%40googlegroups.com.

Azzeddine Remmal

unread,

Dec 12, 2025, 9:23:14 AM (7 days ago) Dec 12

to The Ring Programming Language

Hello Bert and Mahmoud

The example is in https://github.com/azzeddine2017/RingML/tree/main/examples/Chess_End_Game

After running the file chess_train_fast.ring Model training

-------------------------------------------------------------------------------------------------

C:\Users\Azzeddine\Desktop\RingML\examples\Chess_End_Game (main -> origin)
λ ring chess_app.ring
==========================================
Chess End-Game Predictor (AI)
==========================================
Loading model from model/chess_model_fast.rdata...
Done.
Model loaded successfully.

------------------------------------------
Enter Board Configuration (or 'exit'):
White King File (a-h): a
White King Rank (1-8): 6
White Rook File (a-h): b
White Rook Rank (1-8): 4
Black King File (a-h): c
Black King Rank (1-8): 7

>>> AI Prediction: fourteen (Confidence: 25%)
------------------------------------------
Enter Board Configuration (or 'exit'):

-----------------------------------------------------------------------------------------------------

The result >>> AI Prediction: fourteen (Confidence: 25%) means the circuit was successfully completed:

Model Loading: The loadWeights function successfully read the file and restored the network structure.

Input Processing: The transform (a->1) and normalize functions successfully converted your input into numbers the model understands.

Inference: The network calculated the probabilities and selected the highest category.

Result Analysis (25%):
25% may seem low, but in an 18-category classification problem:
Random guessing has a probability of 1/18 ≈ 5.5%.

A 25% result means the model is 5 times smarter than random guessing, which is excellent for a "Fast Mode" model that has only trained for 50 cycles. To reach 80% or 90% accuracy, the model needs longer training (e.g., 5000 cycles), as we did in the XOR example.

In conclusion: RingML is now a fully functional and efficient library.

✅ The Tensor Core is working.

✅ The Layers are working.

✅ Backprop is working.

✅ Data Loading is working.

✅ I/O is working.

✅ The Console App is working.

Regards, Azzeddine

Bert Mariani

unread,

Dec 12, 2025, 10:09:57 AM (7 days ago) Dec 12

to The Ring Programming Language

Hello Mahmoud, Azzeddine

>> " MandelbrotAnimate.ring # Animation does Not Work"

Mea Culpa !

It actually WORKS.

I guess it was not as "Intuitive" as I thought.

Even I forgot what to do as a User !

Luckily I have Comments in the Code.

Because 6 months later I wouldn't have a Clue to What I was Doing.

Regards

Bert Mariani

unread,

Dec 12, 2025, 10:37:51 AM (7 days ago) Dec 12

to The Ring Programming Language

Hello Azzeddine

Chess_End_Game

I downloaded these files to my directory

They refer to --- load "../../src/ringml.ring" which look'ss like a Changes file

Error (E9) : Can't open file ../../src/ringml.ring

Is there a "clean" file that you can include with the rest ?

chess_app.ring
chess_dataset.ring
chess_train_batch.ring
chess_train_fast.ring
chess_utils.ring

Azzeddine Remmal

unread,

Dec 12, 2025, 11:01:09 AM (7 days ago) Dec 12

to The Ring Programming Language

Hello Bert

Hello Bert

The entire library repository needs to be downloaded
https://github.com/azzeddine2017/RingML.git

Bert Mariani

unread,

Dec 12, 2025, 12:54:59 PM (7 days ago) Dec 12

to The Ring Programming Language

Thanks Azzeddine

>>> The entire library repository needs to be downloaded
>>> https://github.com/azzeddine2017/RingML.git

After entering the positions of the White King and Rook, and Black King

Does it Show the Moves as it solves the puzzle ?

==========================================
Chess End-Game Predictor (AI)
==========================================
Loading model from model/chess_model_fast.rdata...
Done.
Model loaded successfully.

------------------------------------------
Enter Board Configuration (or 'exit'):

White King File (a-h): c
White King Rank (1-8): 4
White Rook File (a-h): d
White Rook Rank (1-8): 3
Black King File (a-h): e
Black King Rank (1-8): 4

>>> AI Prediction: fourteen (Confidence: 22%)

------------------------------------------
Enter Board Configuration (or 'exit'):
White King File (a-h):

=============================

WK c4 WR d3 BK e4

Mahmoud Fayed

unread,

Dec 12, 2025, 11:32:53 PM (6 days ago) Dec 12

to The Ring Programming Language

Hello Azzeddine

Thank you very much for adding the Chess End Game sample

In practice we could have

* Say 70% of data used for Training the model

* And 30% of data used for Testing the model (Accuracy Evaluation)

Greetings,

Mahmoud

Mansour Ayouni

unread,

Dec 13, 2025, 1:52:33 PM (6 days ago) Dec 13

to Mahmoud Fayed, The Ring Programming Language

Hello Azzeddine,

Thank you very much for this library, a critical piece of modern ML-based software that Ring developers can finally embrace...

Many thanks to Mahmoud and Bert for their work on the FastoPro extension that made all this possible.

I'll study the library to consider its adoption in implementing the Zai module in Softanza.

All the best,

Mansour

To view this discussion visit https://groups.google.com/d/msgid/ring-lang/5e301bff-c7ee-4d56-922b-35ded548d254n%40googlegroups.com.

Azzeddine Remmal

unread,

Dec 14, 2025, 8:34:55 PM (4 days ago) Dec 14

to The Ring Programming Language

Hello Mahmoud and everyone

The library has been prepared for installation via the package manager.

ringpm install ringml from Azzeddine2017

For a quick training experience as an example of learning the model: `samples/UsingRingML/fast_viz_demo.ring`

For less powerful machines: `samples/UsingRingML/Chess_End_Game/chess_train_lite.ring`

I tried training the following examples:

`samples/UsingRingML/Chess_End_Game/chess_train_split.ring`
`samples/UsingRingML/mnist/mnist_train_split.ring`
But my machine couldn't. The important thing is that the library works efficiently.

RingML - Machine Learning Library for Ring
Version : 1.0.7
Developer : Azzeddine Remmal
License : MIT License

=========================================
QUICK START GUIDE
=========================================
1. Data Preparation:
Use DataSplitter to handle raw CSV data and DataLoader for batching.
see samples/UsingRingML/loader_demo.ring

2. Building the Model:
Construct a model using Tanh for hidden layers and Dropout for regularization.
model = new Sequential
model.add(new Dense(10, 64))
model.add(new Tanh)
model.add(new Dropout(0.2))
model.add(new Dense(64, 3))
model.add(new Softmax)

3. Training:
Use Adam or SGD optimizers.
see samples/UsingRingML/mnist/mnist_train.ring

=========================================
EXAMPLES
=========================================
Find ready-to-run examples in the 'samples/UsingRingML' folder:
- samples/UsingRingML/xor_train.ring : Binary Classification (XOR)
- samples/UsingRingML/classify_demo.ring : Multi-Class Classification
- samples/UsingRingML/mnist/mnist_train.ring : MNIST Digit Recognition
- samples/UsingRingML/Chess_End_Game/ : Full Real-world Project

Regards, Azzeddine

Mahmoud Fayed

unread,

Dec 14, 2025, 9:53:16 PM (4 days ago) Dec 14

to The Ring Programming Language

Hello Azzeddine

Thank you very much for this beautiful package and nice examples :D

Added to Ring Website - News Section :D

Greetings,

Mahmoud

Mahmoud Fayed

unread,

Dec 14, 2025, 10:27:20 PM (4 days ago) Dec 14

to The Ring Programming Language

Hello Azzeddine

When using ring chess_train_fast.ring

I get the next output/error

=== RingML Chess Training (Fast Mode) ===
Reading CSV...
Loaded 28056 games.
Batches per epoch: 110

_________________________________________________________________
Layer (Type) Output Shape Param #
=================================================================
dense (None, 32) 224
_________________________________________________________________
sigmoid (None, 32) 0
_________________________________________________________________
dense (None, 16) 528
_________________________________________________________________
sigmoid (None, 16) 0
_________________________________________________________________
dense (None, 18) 306
_________________________________________________________________
softmax (None, 18) 0
_________________________________________________________________

Total params: 1058
Trainable params: 1058
Non-trainable params: 0
_________________________________________________________________

Starting Training...
......................
Epoch 1 Avg Loss: 1.72251370
......................
Epoch 2 Avg Loss: 2.04692358
......................
Epoch 3 Avg Loss: 2.12166673
......................
Epoch 4 Avg Loss: 2.16214307
......................
Epoch 5 Avg Loss: 2.18348260
......................
Epoch 6 Avg Loss: 2.23748973
......................
Epoch 7 Avg Loss: 2.25608887
......................
Epoch 8 Avg Loss: 2.25293796
......................
Epoch 9 Avg Loss: 2.27011994
......................
Epoch 10 Avg Loss: 2.30402476
......................
Epoch 11 Avg Loss: 2.30986652
......................
Epoch 12 Avg Loss: 2.32623387
......................
Epoch 13 Avg Loss: 2.34301951
......................
Epoch 14 Avg Loss: 2.34176088
......................
Epoch 15 Avg Loss: 2.35730617
......................
Epoch 16 Avg Loss: 2.36630873
......................
Epoch 17 Avg Loss: 2.37230284
......................
Epoch 18 Avg Loss: 2.38214187
......................
Epoch 19 Avg Loss: 2.38145684
......................
Epoch 20 Avg Loss: 2.36868327
......................
Epoch 21 Avg Loss: 2.38010507
......................
Epoch 22 Avg Loss: 2.37716640
......................
Epoch 23 Avg Loss: 2.37919581
......................
Epoch 24 Avg Loss: 2.37894229
......................
Epoch 25 Avg Loss: 2.37585832
......................
Epoch 26 Avg Loss: 2.37202425
......................
Epoch 27 Avg Loss: 2.36944439
......................
Epoch 28 Avg Loss: 2.36674565
......................
Epoch 29 Avg Loss: 2.36578356
......................
Epoch 30 Avg Loss: 2.36504865
......................
Epoch 31 Avg Loss: 2.36238204
......................
Epoch 32 Avg Loss: 2.35935486
......................
Epoch 33 Avg Loss: 2.35653987
......................
Epoch 34 Avg Loss: 2.35408858
......................
Epoch 35 Avg Loss: 2.35194638
......................
Epoch 36 Avg Loss: 2.34993758
......................
Epoch 37 Avg Loss: 2.34796866
......................
Epoch 38 Avg Loss: 2.34604533
......................
Epoch 39 Avg Loss: 2.34419642
......................
Epoch 40 Avg Loss: 2.34244406
......................
Epoch 41 Avg Loss: 2.34082370
......................
Epoch 42 Avg Loss: 2.33937032
......................
Epoch 43 Avg Loss: 2.33806825
......................
Epoch 44 Avg Loss: 2.33685277
......................
Epoch 45 Avg Loss: 2.33567102
......................
Epoch 46 Avg Loss: 2.33451900
......................
Epoch 47 Avg Loss: 2.33342582
......................
Epoch 48 Avg Loss: 2.33242187
......................
Epoch 49 Avg Loss: 2.33151963
......................
Epoch 50 Avg Loss: 2.33071149
Total Time: 1262.94600000s
Saving model to model/chess_model_fast.rdata...

Line 60 Error (R35) : Can't create/open the file
In write() In method saveweights() in file b:\ring\libraries\ringml\src\model/sequential.ring

Called from line 88 in file chess_train_fast.ring

Greetings,

Mahmoud

Mahmoud Fayed

unread,

Dec 14, 2025, 10:49:26 PM (4 days ago) Dec 14

to The Ring Programming Language

Hello Azzeddine

Trying the next sample, stop at this step for many long time!

Greetings,

Mahmoud

Azzeddine Remmal

unread,

Dec 15, 2025, 1:36:00 AM (4 days ago) Dec 15

to The Ring Programming Language

Hello Mahmoud
The reason for the error is that the model folder is missing because I left it empty when I deleted the weight files: `samples\UsingRingML\Chess_End_Game\model`

and `samples\UsingRingML\mnist\model`.

The empty folder is ignored during push.

As for the mnist_train.ring training, it's not stuck. The data is large and requires more processing. Also,

I processed some functions with regular loops in Ring so I could process them in FastPro:

`updateList(aData, :sub, :matrix, oTensor.aData)`. Their behavior is incorrect, so I had to write them in Ring.

`tanh_prime` is missing from FastPro.

I trained the mnist_train.ring model for one epoch after reducing the data to a quarter and resizing the model to a quarter, and the result...

λ ring mnist_app.ring
=== RingML MNIST Digit Predictor ===
Loading Data...
Loaded 10000 test images.
Loading model from model/mnist_model.rdata...
Done.

Enter Image Index (1-10000) or 'exit': 550

Image 550:

= * % % % % % + + :
- % % % % % % % % % % * .
# % % % % % # % % % % % *
% % % = - : = = % % *
* % . . # % =
= = % *
# *
# % .
# % =
. % % =
= % % =
= : : % % % -
= % % % # + : . = % % % %
* % % % % % % % * = . + % % % % :
% % % + * % % % % % % % % % % -
: % % - * % % @ % % % % @ % =
# % : - * % % % % % % % % % % *
* % % % % % % % % % = : = : = * - .
: % % % % % % % *
: * % % * - .

--------------------------
Real Label: 2
AI Predict: 0 (Confidence: 19%)
RESULT: WRONG ❌

Enter Image Index (1-10000) or 'exit': 6624

Image 6624:

# +
- % #
+ : :
% :
. # =
. .
+ % =
% @ :
. % #
- % -
. #
: # % *
. = = +
: * -
+ % +
: @ % -
= % *
* % =
+ % # .
* # :

--------------------------
Real Label: 1
AI Predict: 1 (Confidence: 23%)
RESULT: CORRECT ✅

Enter Image Index (1-10000) or 'exit': 100

Image 100:

. = % @ @ % =
: * % % % % % % % +
* % % % % % % % % % -
= % % % # - . + % % *
% % % = # % % =
% % = : % % #
: % % # % % % +
= % % . * % % % % %
: % % = . * % % % % % +
% % # - . + % % % % % % % :
= % % % % % % % % # % % *
- * % % % % * : % % -
. - : # %
# #
: % =
+ % *
+ % % .
+ % % :
+ % % :
- % % .

--------------------------
Real Label: 9
AI Predict: 9 (Confidence: 19%)
RESULT: CORRECT ✅

Enter Image Index (1-10000) or 'exit': 5000

Image 5000:

: @ @ @ * = = =
: @ @ @ @ @ @ @ @ @ = :
= @ @ @ @ @ @ @ @ @ @ @ @ =
: * @ @ @ @ @ @ @ @ @ @ @ @ @ * :
* @ @ @ = * = * @ @ @ @ @ @ @ @ @ *
: @ @ @ * = * @ @ @ @ @ @ @
= @ @ @ = @ @ @
@ @ @ @ @ @ @
: @ @ @ * = @ @
= @ @ @ : @ @ @
= @ @ @ @ @ @
@ @ @ @ : @ @ @
@ @ @ @ * @ @ @
@ @ @ @ : * @ @ @ @
= @ @ @ * : @ @ @ @ @ :
= @ @ @ @ : : = @ @ @ @ @ @ :
= @ @ @ @ @ @ @ @ @ @ @ @ @ @ @ :
@ @ @ @ @ @ @ @ @ @ @ @ @ = :
= @ * @ @ @ @ @ @ @ @ @ :
: = = * @ = = :

--------------------------
Real Label: 0
AI Predict: 0 (Confidence: 26%)
RESULT: CORRECT ✅

Enter Image Index (1-10000) or 'exit': 33

Image 33:

- % % % % # = .
* % % % % % % % =
: % % # * * . - % % -
+ * : % % -
: % % -
+ % +
: % % % -
- % % % % .
. % % % % % % # -
. % % % % % % % % # .
# # # : : : * % % # .
: + % % # .
* % % :
* % # .
- % % :
. % % % :
: = # % % % :
. . + * % % % % # -
. + % % % % % % % -
= * % % % % = -

--------------------------
Real Label: 3
AI Predict: 3 (Confidence: 20%)
RESULT: CORRECT ✅

Enter Image Index (1-10000) or 'exit': 1

Image 1:

: * + + : .
# % % % % % * * * * * * * * * .
: = : = + % % # % % % % % % % =
: : : : : % % -
: % #
% @ :
= % % .
: % % :
= % *
# % :
= % *
: % % :
# % +
# % # .
. % % :
. # % =
= % % .
: % % % .
= % % # .
= % #

--------------------------
Real Label: 7
AI Predict: 7 (Confidence: 22%)
RESULT: CORRECT ✅

Enter Image Index (1-10000) or 'exit': 3

Image 3:

. % -
- % :
= %
. % +
: % :
# #
. % #
- % *
= % :
: % #
= @ +
* % :
% #
= % +
+ % +
% % :
: % % :
= % #
# % =
* *

--------------------------
Real Label: 1
AI Predict: 1 (Confidence: 26%)
RESULT: CORRECT ✅

Enter Image Index (1-10000) or 'exit': 5000

Image 5000:

: @ @ @ * = = =
: @ @ @ @ @ @ @ @ @ = :
= @ @ @ @ @ @ @ @ @ @ @ @ =
: * @ @ @ @ @ @ @ @ @ @ @ @ @ * :
* @ @ @ = * = * @ @ @ @ @ @ @ @ @ *
: @ @ @ * = * @ @ @ @ @ @ @
= @ @ @ = @ @ @
@ @ @ @ @ @ @
: @ @ @ * = @ @
= @ @ @ : @ @ @
= @ @ @ @ @ @
@ @ @ @ : @ @ @
@ @ @ @ * @ @ @
@ @ @ @ : * @ @ @ @
= @ @ @ * : @ @ @ @ @ :
= @ @ @ @ : : = @ @ @ @ @ @ :
= @ @ @ @ @ @ @ @ @ @ @ @ @ @ @ :
@ @ @ @ @ @ @ @ @ @ @ @ @ = :
= @ * @ @ @ @ @ @ @ @ @ :
: = = * @ = = :

--------------------------
Real Label: 0
AI Predict: 0 (Confidence: 26%)
RESULT: CORRECT ✅

Enter Image Index (1-10000) or 'exit': 444

Image 444:

# * + .
* % + % *
. % + . % =
- % : * *
% + . *
. % : . #
- # . %
+ * . % .
# = %
% . . %
. % . . #
. # : *
. # = *
- # * +
- # . % .
: # * =
% = #
@ : = % =
% # # % =
. % * -

--------------------------
Real Label: 0
AI Predict: 8 (Confidence: 14%)
RESULT: WRONG ❌

Enter Image Index (1-10000) or 'exit': 1000

Image 1000:

= % % % * .
= % # + * % % .
= @ * : % % .
% # : % # =
: % - + % %
= % = % #
# * = % *
* * = % -
: % # : = % - =
= % % : = % + # +
- # % = # % % #
= % % % # % % #
= = = % %
# %
# *
: % -
* % .
# %
: % =
: % .

--------------------------
Real Label: 9
AI Predict: 3 (Confidence: 15%)
RESULT: WRONG ❌

Enter Image Index (1-10000) or 'exit': exit

ringml updated to 1.0.8

ringpm update ringml

Regards, Azzeddine

Mahmoud Fayed

unread,

Dec 15, 2025, 2:08:15 AM (4 days ago) Dec 15

to The Ring Programming Language

Hello Azzeddine

Thank you very much for the update :D

(1) If you discovered an issue in RingFastPro, send a pull request with the fix or necessary additions
Updating the extension to match RingML requirements is a good idea to improve RingFastPro

(2) Revise the examples section in the output, to include all of the beautiful samples :D

=========================================
EXAMPLES
=========================================
Find ready-to-run examples in the 'samples/UsingRingML' folder:
- samples/UsingRingML/xor_train.ring : Binary Classification (XOR)
- samples/UsingRingML/classify_demo.ring : Multi-Class Classification
- samples/UsingRingML/mnist/mnist_train.ring : MNIST Digit Recognition
- samples/UsingRingML/Chess_End_Game/ : Full Real-world Project

Missing samples are:

* classify_demo2.ring

* fast_viz_demo.ring

Keep up the GREAT WORK :D

Greetings,

Mahmoud

Azzeddine Remmal

unread,

Dec 15, 2025, 6:40:10 PM (3 days ago) Dec 15

to The Ring Programming Language

Hello Mahmoud and everyone,

As you know, FastPro is a large, complex public library.

Every change requires monitoring and verification of its effects on related applications.

This slows down development, so I created a RingTensor extension

specifically for the RingML library. https://github.com/azzeddine2017/ringtensor

I preserved the old RingML repository using FastPro. https://github.com/azzeddine2017/ringml

And I created a new RingML repository using RingTensor. https://github.com/azzeddine2017/ringml-using-ringtensor

After some testing, the speeds are similar, and I'm working on identifying the bottleneck—the exact cause of the slow training.

Regards, Azzeddine

Message has been deleted

Mahmoud Fayed

unread,

Dec 15, 2025, 9:55:53 PM (3 days ago) Dec 15

to The Ring Programming Language

Hello Azzeddine

>> "Every change requires monitoring and verification of its effects on related applications."

We can automate the tests, as we have here: ring/language/tests at master · ring-lang/ring

>> "After some testing, the speeds are similar, and I'm working on identifying the bottleneck—the exact cause of the slow training"

RingFastPro is already fast C code, to improve the performance, you need a different direction

(1) Different algorithm/parameters/dataset

(2) Using Threads

(3) Using GPU: CUDA Programming Guide — CUDA Programming Guide

Suggestion:

What about creating an extension for mlpack: mlpack - Home

It's a great library/tool (command line), easy to use and comes with many algorithms

If you created an extension for such a great library, it will be very useful/practical, and could be used in many apps

Also, since the c/c++ code is already written by mlpack team, we can focus on usability, and later building a DSL on top of RingMLPACK, so we win the best of both worlds.

Keep up the GREAT WORK :D

Greetings,

Mahmoud

Mansour Ayouni

unread,

Dec 16, 2025, 4:57:52 AM (3 days ago) Dec 16

to Mahmoud Fayed, The Ring Programming Language

Hello Azzeddine,

I salute your work and your strong sense of improvement and achievement. This project has clearly helped you gain a deeper understanding of core machine learning requirements, and its educational value is unquestionable.

To move toward a professional-grade ML solution that we can confidently use in real-world, customer-facing projects, I recommend considering Mahmoud’s proposal: developing an extension based on mlpack. This approach leverages a robust ecosystem of well-established libraries—Armadillo for linear algebra, ensmallen for nonlinear optimization, and cereal for data serialization—widely adopted in mature ML suites.

Such an extension would represent a significant and valuable contribution to the Ring ecosystem’s machine learning capabilities.

All the best,

Mansour

To view this discussion visit https://groups.google.com/d/msgid/ring-lang/57b907e5-d7e3-4bc5-a13f-c2afbbff3845n%40googlegroups.com.

Azzeddine Remmal

unread,

Dec 16, 2025, 11:01:42 AM (3 days ago) Dec 16

to The Ring Programming Language

Hello Mahmoud and Mansour

Thank you for the valuable advice.

I haven't slept since yesterday until I found the solution.

As you know, Mahmoud, perseverance always pays off, and this is a detailed report.

Performance Optimization Report: How I Fixed Slow Training in RingML

1. The Problem: The "Overhead" Bottleneck

Initially, the library suffered from extreme slowness during training (e.g., taking ~60 seconds for a single epoch on medium data). Despite using a C-extension (FastPro), the CPU usage remained low (~30%), indicating the processor was waiting rather than computing.

Root Causes identified via Benchmarking:

Data Marshalling Overhead: Every mathematical operation (e.g., Add, MatMul) required copying data from Ring Lists to C Arrays and back to Ring Lists. This copy operation consumed 95% of the execution time, while the actual math took only 5%.

Memory Churn (GC Pressure): The library was creating new Lists for every intermediate calculation in every layer. This forced the Ring Garbage Collector to work constantly to clean up memory, causing massive latency.

Naive C Algorithms: The initial Matrix Multiplication implementation used standard i-j-k loops, which caused frequent CPU Cache Misses because memory was being accessed non-sequentially.

Interpreter Loops: Complex logic like the Adam Optimizer was implemented using Ring loops. Updating 10,000 weights meant executing ~80,000 interpreter instructions per batch.

2. The Solution: RingTensor & Memory-Resident Architecture

To solve these issues, I completely re-engineered the core engine by building a custom C-extension named RingTensor.

A. Architecture Shift: Memory-Resident Tensors (Zero-Copy)

Old Approach: Tensor object held a Ring List (aData).
New Approach: Tensor object holds a C Pointer (pData) pointing to a raw double* array in RAM.
Impact: Ring no longer "sees" or copies the data. It only passes the memory address to C functions. This eliminated the data marshalling overhead completely.

B. Algorithmic Optimization: Cache-Friendly Math

Action: I rewrote the Matrix Multiplication (MatMul) logic in C.
Technique: Switched from i-j-k loop order to i-k-j loop order.
Result: This allows the CPU to access memory sequentially (Linear Access Pattern), maximizing SIMD usage and minimizing Cache Misses.

C. Fused Kernels (The Speed Booster)

Action: Instead of calling separate functions for Add, Mul, Div, and Sqrt during weight updates.
Technique: I implemented "Fused Kernels" for Optimizers (Adam, SGD) inside C.
Result: The entire update logic (Momentum, Velocity, Bias correction, Weight update) happens in a single C function call per layer, bypassing the interpreter entirely.

D. Zero Allocation Strategy

Action: I modified Dense layers to pre-allocate memory for outputs and gradients during initialization (init).
Result: During training loops, the C functions now write results directly into these pre-existing buffers instead of allocating new memory. Garbage Collection time dropped to ~0.005 seconds.

E. Compiler Optimization

Action: Adjusted build flags to use /O2 (Maximize Speed) and /LTCG (Link Time Code Generation).

3. The Result

By moving from "Safe Mode" (Ring Loops) to "Turbo Mode" (Memory-Resident C Pointers), I achieved:

Speedup: ~100x faster training.
Memory: Stable consumption with no leaks.
Accuracy: Full Double-Precision (64-bit) math ensured convergence where previous versions failed.

RingML is now a true High-Performance Deep Learning Library.

Regards, Azzeddine

Bert Mariani

unread,

Dec 16, 2025, 11:22:55 AM (3 days ago) Dec 16

to The Ring Programming Language

Hello Azzeddine

Great Observations and Measurements, Solutions and Re-Writes !!

Yes, Contiguous Memory Arrays are Best on the Hardware execution side.

Less Churn and Cache Misses. Hardware works as it was Designed to Run.

Use of Pointers avoids all the Copying Overhead, Allocation Overhead etc..

100x Speed Improvement, shows what can be done.

Regards

Bert Mariani

Mahmoud Fayed

unread,

Dec 16, 2025, 12:03:55 PM (3 days ago) Dec 16

to The Ring Programming Language

Hello Azzeddine

>> " Initially, the library suffered from extreme slowness during training (e.g., taking ~60 seconds for a single epoch on medium data). Despite using a C-extension (FastPro), the CPU usage remained low (~30%), indicating the processor was waiting rather than computing."

>> " Every mathematical operation (e.g., Add, MatMul) required copying data from Ring Lists to C Arrays and back to Ring Lists. This copy operation consumed 95% of the execution time, while the actual math took only 5%."

What you need is fixing your Ring code, not creating another library in C! because RingFastPro is already written in C

RingFastPro is fast, and Image Pixel is already using it to process images at high performance

Yes, it can be faster, but the real problem is not here

The problem is using unnecessary assignments like this in RingML: ringml/libraries/ringml/src/core/tensor.ring at main · azzeddine2017/ringml

aData = updateList(aData, :add, :items, nVal)

You don't need aData =

Because updateList() function already update aData directly to avoid the copy operations

And using updateList() you can do many operations at once

Documentation: Using FastPro — Ring 1.24.0 documentation

Please check RingFastPro samples and see how it's used.

Greetings,

Mahmoud

Azzeddine Remmal

unread,

Dec 16, 2025, 1:53:28 PM (3 days ago) Dec 16

to The Ring Programming Language

Hello Bert, Mahmoud, and Mansour,

Thank you all for this enriching discussion. I am truly grateful for your combined insights, which cover everything from low-level hardware execution to high-level ecosystem strategy.

To Bert:
Thank you for the validation regarding the memory model. You were spot on about contiguous memory arrays. My profiling confirmed that the shift from Ring Lists—which offer great flexibility but can be scattered in memory—to raw C pointers was the decisive factor. As you noted, minimizing the copying overhead and reducing CPU cache misses allowed the hardware to execute the math exactly as designed, resulting in the massive speed improvement I observed.

To Mansour:
I appreciate your kind words and your recognition of the educational value of this project. You are absolutely right; building RingML from scratch provided me with a profound understanding of the core requirements needed to sustain a machine learning workload. I agree that for mission-critical, customer-facing projects, leveraging an established ecosystem like mlpack (with its reliance on Armadillo and ensmallen) is the pragmatic path forward for the Ring community.

To Dr. Mahmoud:
Thank you for the code review and the strategic suggestions. I have corrected the redundant assignment in the Ring code as you pointed out.
Regarding the suggestion to wrap mlpack: I fully agree with you and Mansour. Creating an extension for mlpack would be the most effective way to bring a production-ready, feature-rich ML suite to Ring, combining the best of both worlds.
However, the specific goal of RingML was to demonstrate that Ring can serve as a powerful "glue language" for a custom-built, high-performance engine, much like how Python interacts with PyTorch. I felt it was necessary to build the engine architecture from the ground up to push the limits of Ring's C-API.
While FastPro is excellent for general list manipulation, I found that Deep Learning workloads have specific constraints that required the specialized RingTensor approach:
Memory Layout: Deep Learning relies heavily on accessing memory in a linear, contiguous block to maximize CPU performance. I moved to raw C pointers to ensure the CPU could prefetch data efficiently, avoiding the overhead of traversing linked structures during heavy matrix operations.
Kernel Fusion: Optimization algorithms involve multiple sequential steps—updating momentum, velocity, and weights based on gradients. Doing this via separate function calls from the interpreter creates significant overhead per step. In my custom extension, I implemented "Fused Kernels," where the entire optimizer logic executes in a single C function call, bypassing the interpreter entirely for the heavy lifting.
RingML now stands as a proof-of-concept that Ring is capable of driving high-performance numerical computing. I look forward to exploring the mlpack integration as the next logical step for the ecosystem.

Best regards,
Azzeddine

Mansour Ayouni

unread,

Dec 16, 2025, 2:04:23 PM (3 days ago) Dec 16

to Azzeddine Remmal, The Ring Programming Language

Hello Azzeddine,

I’m glad to see how strongly you believe in your direction of building a Ring-based ML stack from scratch. I’m confident that, in the end, you’ll deliver a solution that is truly well-suited to the specific needs of Ring applications, even in a professional context.

I encourage you to keep moving forward, to study and learn from existing libraries such as mlpack and others, and to incorporate their strengths in a way that preserves your own vision of RingML—clean, elegant, and practical.

I would be very happy if someone develops an mlpack extension for Ring, but I would be just as happy to see you continue your journey with RingML itself.

All the best,

Mansour

To view this discussion visit https://groups.google.com/d/msgid/ring-lang/51a6b7b4-47d4-4380-9e92-61d1be7c1fben%40googlegroups.com.

Mahmoud Fayed

unread,

Dec 16, 2025, 2:08:43 PM (3 days ago) Dec 16

to The Ring Programming Language

Hello Azzeddine

>> "To Dr. Mahmoud:"

No need to say Dr. before my name

In the group we are contributing to the same project, and we call each one with his name directly

And inside our heart, we love/respect each member in the group as a friend/brother/colleage

>> "I have corrected the redundant assignment in the Ring code as you pointed out."

Thank you very much for the update :D

>> "In my custom extension, I implemented "Fused Kernels," where the entire optimizer logic executes in a single C function call, bypassing the interpreter entirely for the heavy lifting."

I remove RingML from Ring website - News Section, and I will test the new faster project that you developed and uses your extension, so we can add it to Ring website as a better solution.

Please add the new extensions to RingPM registry

>> "I look forward to exploring the mlpack integration as the next logical step for the ecosystem."

Thank you very much for investigating this direction, this will be a GREAT CONTRIBUTION

Keep up the GREAT WORK :D

Greetings,

Mahmoud

Azzeddine Remmal

unread,

Dec 16, 2025, 8:48:50 PM (2 days ago) Dec 16

to The Ring Programming Language

Hello Bert

♟️ RingML Chess AI Engine: User Guide

This application demonstrates the power of RingML by turning a trained Neural Network into a functional Chess Engine. It doesn't just classify positions; it actively searches for the best move to checkmate the Black King.

📋 Prerequisites

Before running the application, ensure you have the following directory structure and files:

Model: The trained model file chess_universal.rdata must exist inside the model/ folder.

Note: This model was trained using the UniversalDataset method with Data Splitting.

Images: A folder named imgs/ containing the piece assets:

WKing.png, WRook.png, BKing.png

Library: The src/ folder of RingML must be accessible.

🚀 How to Run

Execute the application from the command line:

download
content_copy
expand_less
ring chess_ai_tool.ring
🎮 How to Use

The application features a graphical interface divided into two sections: Control Panel (Left) and Chess Board (Right).

1. Setup the Board

Use the dropdown menus on the left to place the three pieces on the board:

♔ White King: Select File (a-h) and Rank (1-8).

♖ White Rook: Select File (a-h) and Rank (1-8).

♚ Black King: Select File (a-h) and Rank (1-8).

The board on the right will update instantly to reflect your setup.

2. The "AI Move" Button

Click the 🤖 AI Move button to let the Neural Network decide the next move.

3. Interpreting Results

The AI performs two actions:

Move Execution: It moves the White piece (King or Rook) on the graphical board.

🟨 Yellow Square: The position the piece moved from.

🟩 Green Square: The position the piece moved to.

Status Report:

Prediction: Displays the expected game outcome (e.g., Mate in Two, Draw).

Confidence: The probability percentage of that outcome.

Log: Displays the text description of the move (e.g., "Best Move: WR to a7").

🧠 How It Works (The Logic)

This is not a traditional chess engine based on Minimax or Alpha-Beta pruning. It is a Neural-Guided Engine:

State Analysis: When you click the button, the app reads the current coordinates of the pieces.

Move Generation: It calculates all legal moves for the White King and White Rook (one step ahead).

Neural Evaluation:

For every possible move, it creates a new "Hypothetical Board State".

It feeds this state into the RingML Tensor and passes it through the loaded neural network (chess_universal.rdata).

Decision Making:

The model predicts how close White is to winning for each move (0 moves = Mate, 16 moves = Far).

The engine selects the move that minimizes the "Distance to Mate" (i.e., looks for the fastest win) while avoiding Draws.

This demonstrates that a generic Neural Network built with RingML can "understand" the rules and strategy of a complex game simply by learning from data.

Best regards,
Azzeddine

Mahmoud Fayed

unread,

Dec 16, 2025, 11:58:17 PM (2 days ago) Dec 16

to The Ring Programming Language

Hello Azzeddine

When trying to install the ringml-using-ringtensor package

I get the following error message

Download File : samples/UsingRingML/Chess_End_Game/chess_train_adam.ring
Can't download the file : https://raw.githubusercontent.com/azzeddine2017/ringml-using-ringtensor/master/samples/UsingRingML/Chess_End_Game/chess_train_adam.ring
Download File : samples/UsingRingML/Chess_End_Game/chess_train_batch.ring
Can't download the file : https://raw.githubusercontent.com/azzeddine2017/ringml-using-ringtensor/master/samples/UsingRingML/Chess_End_Game/chess_train_batch.ring
Download File : samples/UsingRingML/Chess_End_Game/chess_train_fast.ring
Can't download the file : https://raw.githubusercontent.com/azzeddine2017/ringml-using-ringtensor/master/samples/UsingRingML/Chess_End_Game/chess_train_fast.ring
Download File : samples/UsingRingML/Chess_End_Game/chess_train_lite.ring
Download File : samples/UsingRingML/Chess_End_Game/chess_train_split.ring
Download File : samples/UsingRingML/Chess_End_Game/chess_utils.ring
Download File : samples/UsingRingML/Chess_End_Game/data/chess.csv
Download File : samples/UsingRingML/Chess_End_Game/model/chess_model_lite.rdata
Download File : samples/UsingRingML/classify_demo.ring
Download File : samples/UsingRingML/classify_demo2.ring
Can't download the file : https://raw.githubusercontent.com/azzeddine2017/ringml-using-ringtensor/master/samples/UsingRingML/classify_demo2.ring

Download File : samples/UsingRingML/xor_train.ring
Preparing ringml-using-ringtensor
Error(3) : We don't have this package
Package Name : ringml

Greetings,

Mahmoud

Bert Mariani

unread,

Dec 17, 2025, 9:46:38 AM (2 days ago) Dec 17

to The Ring Programming Language

Hello Azzeddine

c:\ring\bin> ringpm install ringml-using-ringtensor from Azzeddine2017

Where are the "images" file etc ?

Same problem as Mahmoud:

....

Download File : libraries/ringml/tests/test_sub.ring
Download File : samples/UsingRingML/Chess_End_Game/chess_app.ring
Download File : samples/UsingRingML/Chess_End_Game/chess_dataset.ring
Download File : samples/UsingRingML/Chess_End_Game/chess_final_model.ring
Download File : samples/UsingRingML/Chess_End_Game/chess_train_adam.ring
Can't download the file : https://raw.githubusercontent.com/Azzeddine2017/ringml-using-ringtensor/master/samples/UsingRingML/Chess_End_Game/chess_train_adam.ring
Download File : samples/UsingRingML/Chess_End_Game/chess_train_batch.ring
Can't download the file : https://raw.githubusercontent.com/Azzeddine2017/ringml-using-ringtensor/master/samples/UsingRingML/Chess_End_Game/chess_train_batch.ring
Download File : samples/UsingRingML/Chess_End_Game/chess_train_fast.ring
Can't download the file : https://raw.githubusercontent.com/Azzeddine2017/ringml-using-ringtensor/master/samples/UsingRingML/Chess_End_Game/chess_train_fast.ring

Download File : samples/UsingRingML/Chess_End_Game/chess_train_lite.ring
Download File : samples/UsingRingML/Chess_End_Game/chess_train_split.ring
Download File : samples/UsingRingML/Chess_End_Game/chess_utils.ring
Download File : samples/UsingRingML/Chess_End_Game/data/chess.csv
Download File : samples/UsingRingML/Chess_End_Game/model/chess_model_lite.rdata
Download File : samples/UsingRingML/classify_demo.ring
Download File : samples/UsingRingML/classify_demo2.ring

Can't download the file : https://raw.githubusercontent.com/Azzeddine2017/ringml-using-ringtensor/master/samples/UsingRingML/classify_demo2.ring
Download File : samples/UsingRingML/fast_viz_demo.ring
Download File : samples/UsingRingML/loader_demo.ring
Download File : samples/UsingRingML/mnist/data/mnist_test.csv
Download File : samples/UsingRingML/mnist/model/mnist_model.rdata

Azzeddine Remmal

unread,

Dec 17, 2025, 11:42:33 AM (2 days ago) Dec 17

to The Ring Programming Language

Hello Bert, Mahmoud,

I was correcting errors everywhere and re-correcting the examples and tests.

Now the library is ready for testing.

I noticed that the package manager doesn't delete folders when removing the package; it removes the files but leaves the folders. As for training the models, it's good, but for good results, the model needs to be adjusted in terms of parameters and the training Epochs need to be increased.

Best regards,
Azzeddine

Azzeddine Remmal

unread,

Dec 17, 2025, 3:10:16 PM (2 days ago) Dec 17

to The Ring Programming Language

Hello Mahmoud

Update the library RingML to version 1.1.1

ringpm update ringml-using-ringtensor

Best regards,
Azzeddine

Bert Mariani

unread,

Dec 17, 2025, 4:58:28 PM (2 days ago) Dec 17

to The Ring Programming Language

Hello Azzeddine

ringpm install ringml-using-ringtensor from Azzeddine2017

Download files - NO errors encountered

Then ran the Samples

Some Error captures are shown below

>>>>>>>>>>>>>>>>>>>>>>

=========================================
RUN AN EXAMPLE
=========================================
cd samples/UsingRingML/
ring xor_train.ring

>>>>>>>>>>>>>>>>>>>>>>>>>>>>>

========================

Error (E9) : Can't open file ../src/ringml.ring

C:/ring/samples/UsingRingML/mnist/mnist_train.ring Line (10) Error (C27) : Syntax Error!

Error (E9) : Can't open file ../src/utils/visualizer.ring

C:/ring/samples/UsingRingML/mnist/mnist_train.ring Line (11) Error (C27) : Syntax Error!

C:/ring/samples/UsingRingML/mnist/mnist_train.ring errors count : 2

============================

=== RingML Multi-Class Classification Demo ===
Training...

Line 34 Error (R3) : Calling Function without definition: variableexists
In method hasattribute() in file C:\ring\libraries\ringml\src\optim/sgd.ring

Called from line 12 In method update() in file C:\ring\libraries\ringml\src\optim/sgd.ring

Called from line 56 in file C:/ring/samples/UsingRingML/classify_demo.ring

====================================

=== RingML XOR Training Example ===

Line 14 Error (R12) : Error in property name, property not found: adata
in file C:/ring/samples/UsingRingML/xor_train.ring

==========================

Azzeddine Remmal

unread,

Dec 17, 2025, 6:53:05 PM (2 days ago) Dec 17

to The Ring Programming Language

Hello Bert

Thank you so much for testing and reviewing the errors.
The error was caused by a function I accidentally deleted: variableexists.

I forgot to update the XOR example. Now all the examples work.
I removed the Color Console library and replaced it with the Styler class. Thanks to Youssef.
I removed the mnist_train.ring example and replaced it with mnist_train_universal.ring to demonstrate the UniversalDataset class's ability to load and manage data.
I would like your opinion on the chess_ai_tool.ring gui example. I developed it because you requested it.

Update the library RingML to version 1.1.2

ringpm update ringml-using-ringtensor

_____________________________________________________________________________________________________________________________
C:\Users\Azzeddine\Desktop\ringml-using-ringtensor\samples\UsingRingML (main -> origin)
λ ring xor_train.ring
=== RingML XOR Training Example (Fixed) ===
Training started for 5000 epochs...
Epoch 500 : Loss = 0.2383298372
Epoch 1000 : Loss = 0.1628673734
Epoch 1500 : Loss = 0.0478212129
Epoch 2000 : Loss = 0.0132908407
Epoch 2500 : Loss = 0.0064522787
Epoch 3000 : Loss = 0.0040387804
Epoch 3500 : Loss = 0.0028729276
Epoch 4000 : Loss = 0.0022027686
Epoch 4500 : Loss = 0.0017733661
Epoch 5000 : Loss = 0.0014771626

=== Training Finished ===
Final Predictions:
Tensor Shape: (4, 1)
| 0.0369000000 |
| 0.9664000000 |
| 0.9574000000 |
| 0.0400000000 |

Expected Targets:
Tensor Shape: (4, 1)
| 0 |
| 1 |
| 1 |
| 0 |

C:\Users\Azzeddine\Desktop\ringml-using-ringtensor\samples\UsingRingML (main -> origin)
λ ring save_load_demo.ring
=== RingML Save/Load Model Demo ===
--> Training Model A...
..........
Model A Predictions (Trained):
Tensor Shape: (4, 1)
| 0.1103 |
| 0.8533 |
| 0.8410 |
| 0.1909 |

Saving model to xor_weights.rdata...
Done.

--> Creating Model B (Same Architecture, Untrained)...
Model B Predictions (Before Loading):
Tensor Shape: (4, 1)
| 0.467799999999999994 |
| 0.478399999999999992 |
| 0.454299999999999982 |
| 0.462600000000000011 |

Loading model from xor_weights.rdata...
Done.
Model B Predictions (After Loading):
Tensor Shape: (4, 1)
| 0.110299999999999995 |
| 0.853299999999999947 |
| 0.840999999999999970 |
| 0.190899999999999986 |

--> Verification: SUCCESS (Models Match)

C:\Users\Azzeddine\Desktop\ringml-using-ringtensor\samples\UsingRingML (main -> origin)
λ ring loader_demo.ring
=== DataLoader Demo (Pointer Based) ===
Tensor Attributes: pdata
nrows
ncols

Total Samples: 10
Batch Size: 4
Num Batches: 3

--- Batch 1 ---
Input Shape: Tensor Shape: (4, 2)
| 1, 2 |
| 2, 4 |
| 3, 6 |
| 4, 8 |

Target Shape: Tensor Shape: (4, 1)
| 1 |
| 1 |
| 1 |
| 1 |

--- Batch 2 ---
Input Shape: Tensor Shape: (4, 2)
| 5, 10 |
| 6, 12 |
| 7, 14 |
| 8, 16 |

Target Shape: Tensor Shape: (4, 1)
| 1 |
| 1 |
| 1 |
| 1 |

--- Batch 3 ---
Input Shape: Tensor Shape: (2, 2)
| 9, 18 |
| 10, 20 |

Target Shape: Tensor Shape: (2, 1)
| 1 |
| 1 |

C:\Users\Azzeddine\Desktop\ringml-using-ringtensor\samples\UsingRingML (main -> origin)
λ ring fast_viz_demo.ring
=== RingML Visual Demo (Guaranteed Learning) ===
Generating distinct data...

_________________________________________________________________
Layer (Type) Output Shape Param #
=================================================================

dense (None, 64) 192
_________________________________________________________________
relu (None, 64) 0
_________________________________________________________________
dense (None, 32) 2080
_________________________________________________________________
relu (None, 32) 0

_________________________________________________________________
dense (None, 16) 528
_________________________________________________________________

relu (None, 16) 0
_________________________________________________________________
dense (None, 3) 51
_________________________________________________________________
softmax (None, 3) 0
_________________________________________________________________

Total params: 2851
Trainable params: 2851
Non-trainable params: 0
_________________________________________________________________

==========================================
RingML v - (1.1.1) Training Dashboard
==========================================
[DONE] Epoch 1: Avg Loss = 3.2026346052 | Val Accuracy = 65.8333333333 %
\ Epoch 2/20 [====================] 100| Loss: 0.3306364060 | Accuracy: 0 %^C
C:\Users\Azzeddine\Desktop\ringml-using-ringtensor\samples\UsingRingML (main -> origin)
λ ring classify_demo.ring

=== RingML Multi-Class Classification Demo ===
Training...

Epoch 500 Loss: 0.00668052
Epoch 1000 Loss: 0.00287809
Epoch 1500 Loss: 0.00179645
Epoch 2000 Loss: 0.00129492
Epoch 2500 Loss: 0.00100789
Epoch 3000 Loss: 0.00082282
=== Final Predictions ===
Tensor Shape: (3, 3)
| 0.99910000, 0.00010000, 0.00060000 |
| 0.00010000, 0.99930000, 0.00040000 |
| 0.00050000, 0.00040000, 0.99890000 |

=== Expected Targets ===
Tensor Shape: (3, 3)
| 1, 0, 0 |
| 0, 1, 0 |
| 0, 0, 1 |

C:\Users\Azzeddine\Desktop\ringml-using-ringtensor\samples\UsingRingML (main -> origin)
λ ring benchmark_bottleneck.ring
=== RingML Performance Benchmark (C-Pointer Mode) ===
Generating Data (1000x1500)... Done in 0.0480s

_________________________________________________________________
Layer (Type) Output Shape Param #
=================================================================

dense (None, 200) 300200
_________________________________________________________________
tanh (None, 200) 0
_________________________________________________________________
dense (None, 10) 2010
_________________________________________________________________
softmax (None, 10) 0
_________________________________________________________________

Total params: 302210
Trainable params: 302210
Non-trainable params: 0
_________________________________________________________________

Running Benchmark for 10 epochs...
Measuring average time per operation...
Epoch 10 done.

=== RESULTS (Total Time: 73.2070s) ===
Forward Pass : 47.6530 s
Loss Calc : 0.8460 s
Backward Pass: 20.2340 s
Optimizer : 4.4480 s
Garbage Coll : 0.0230 s

_________________________________________________________________

Best regards,
Azzeddine

Bert Mariani

unread,

Dec 18, 2025, 10:33:14 AM (18 hours ago) Dec 18

to The Ring Programming Language

Hello Azzeddine

ingpm update ringml-using-ringtensor

==============================

Chess: chess_ai_tool.ring

See 2 image captures below

=============================

=== RingML Multi-Class Classification Demo ===

Line 13 Error (R12) : Error in property name, property not found: adata
in file C:/ring/samples/UsingRingML/classify_demo2.ring

==============================

=== RingML MNIST Digit Predictor ===

Loading Data (data/mnist_test_lite.csv)...

Line 26 Bad parameter type!
In lower() in file C:/ring/samples/UsingRingML/mnist/mnist_app.ring

=================================

Error (E9) : Can't open file ../src/ringml.ring

C:/ring/samples/UsingRingML/mnist/mnist_train.ring Line (10) Error (C27) : Syntax Error!

Error (E9) : Can't open file ../src/utils/visualizer.ring

C:/ring/samples/UsingRingML/mnist/mnist_train.ring Line (11) Error (C27) : Syntax Error!

C:/ring/samples/UsingRingML/mnist/mnist_train.ring errors count : 2

===============================

Image Capture 1

WK next to BK is invalid

BK never moves ?

Image Capture 2

Initial position: WR b6

Moves are always repetive as shown

No progress

BK never moves

Reply all

Reply to author

Forward