Creation of SB-Neuron. Ours. Branded.(v2)

[litdev]

Go for it!

[AbsoluteBeginner]

(translated by Google translator)

Friends,
today is a significant day for every fan of our wonderful Small Basic. 
I think I have everything ready to create a demo program that will show everyone the incredible capabilities of our proprietary SB-Neuron.

Thanks to LitDev and all the people who participated previously and are now participating in improving the SB programming system, I got the opportunity to create our incredible SB Neuron without having to learn any programming language “for adults”.

In my future demo program, one SB-Neuron will gain the ability to very accurately determine the value of the "H" parameter in the HSL color format, having non-normalized RGB data at the input.
Using a description of just one point and 48 Bezier curves, this neuron will be able to give accurate answers to questions about the value of "H" at 65,536 points in the RGB range, starting from (112,0,0) and ending at (112,255,255).
(a screenshot of the graph of this range section is in my OneDrive with the name "RGB to H for SB-Neuron 04.jpg". https://1drv.ms/f/s!AnoSlTzMqlL6jNx6H4Qi...w?e=8D73FE )

I would like to specifically draw your attention to the fact that in addition to the incredible ability of the SB-Neuron to describe such a complex graph, the SB-Neuron DOES NOT REQUIRE SPECIAL PREPARATION of the input data. The output data of the neuron is also immediately ready for use. Without processing.

These are the miracles that our simplest, most convenient and skillful Small Basic allows ordinary people to do. 

[AbsoluteBeginner]

(translated by Google translator)

Hi all.
On Thursday and Friday I failed to write a single letter of the code. 
But on Saturday and today I managed to teach my program to detect the beginning and end of a range of training data that can be described by a single Bezier curve.
Now I teach my program to select the parameters of the curve so that the curve matches the real data quite accurately.

What's new with you? 

[AbsoluteBeginner]

(translated by Google translator)

Hi all. 
Here I am writing, to the best of my ability, the code of our SB-Neuron and thinking: at the moment, I can’t imagine a problem that SB-Neuron can’t solve.
Even now, the SB-Neural Network is capable of studying an array of completely random data in order to then reproduce the required answers during the use of this neural network. ( If I understand everything correctly, then a regular neural network cannot learn on random data. )
On ordered data, SB-Neuron learns very quickly and with very high accuracy.

Therefore, I ask you to come up with some difficult problem that our SB-Neuron cannot solve.
The complexity of the task should not be the need to process a HUGE AMOUNT of data.
After I finish my current demo program, I will try to solve your test problem.

We agreed?.. 

[jrmrhrb00]

AB,

Have your neuron system be able to detect Boolean or, and, not, nand, Xor, and Xnor mixed in the data. I can do it one at a time, but I haven't figured out how to show and display multiple detections.

JR

[AbsoluteBeginner]

AB,

Have your neuron system be able to detect Boolean or, and, not, nand, Xor, and Xnor mixed in the data. I can do it one at a time, but I haven't figured out how to show and display multiple detections.

JR

Hello, JR. 
Thank you for agreeing to help me.
I want to publish my demo program tonight. After that, I can start working on solving your problem.

But, at the moment, I do not understand the essence of this task.
Can you send me a small sample of the training data so I can understand the gist of the problem?

Thank you. 

[jrmrhrb00]

AB,

I will attach a file for an or gate. I haven't figured out how to do multiple gates. I couldn't do an attachment. So, I put it in as code/

JR

Code:

inputNode = 2
hiddenNode = 2
outputNode = 1
'scale = 100
trainingDataText = Program.Directory+"\orANNInput.txt"
trainingDataBinary = Program.Directory+"\orANNInput.bin"

name= "orANN"
NeuralNetwork.New(name,inputNode+","+hiddenNode+","+outputNode)
NeuralNetwork.BinaryOutput(name,0,"true")
'name = NeuralNetwork.Load(Program.Directory+"\"+name+".textwindow")

File.DeleteFile(trainingDataText)

separationCharacter = " " 'Can be LDText.LF, LDText.TAB, " " or ","

'orann
File.AppendContents(trainingDataText,"0 0 0")
File.AppendContents(trainingDataText,"1 0 1")
File.AppendContents(trainingDataText,"0 1 1")
File.AppendContents(trainingDataText,"1 1 1")

NeuralNetwork.ConvertFile(trainingDataText,trainingDatabinary,"False")
textwindow.writeline("Training started")
epoch = NeuralNetwork.Train(name,trainingDataBinary,"True")
NeuralNetwork.Save(name,Program.Directory+"\"+name+".txt")
textwindow.writeline("epoch="+epoch)
textwindow.writeline("Trained="+NeuralNetwork.Trained(name))
textwindow.writeline("BinaryOutput="+NeuralNetwork.BinaryOutput(name,0,"True"))
textwindow.writeline("Epoch="+NeuralNetwork.Epochs(name,0,"True"))
textwindow.writeline("LearningRate="+NeuralNetwork.LearningRate(name,0,"True"))
textwindow.writeline("Momentum="+NeuralNetwork.Momentum(name,0,"True"))
textwindow.writeline("SigmoidResponse="+NeuralNetwork.SigmoidResponse(name,0,"True"))
textwindow.writeline("ErrorRequired="+NeuralNetwork.ErrorRequired(name,0,"True"))

input = 0+","+0
output=NeuralNetwork.use(name,input)
textwindow.writeline(input)
textwindow.writeline(output)

input=1+","+0
output=NeuralNetwork.Use(name,input)
textwindow.writeline(input)
textwindow.writeline(output)

input=0+","+1
output=NeuralNetwork.Use(name,input)
textwindow.writeline(input)
textwindow.writeline(output)

input=1+","+1
output=NeuralNetwork.Use(name,input)
textwindow.writeline(input)
textwindow.writeline(output)

[AbsoluteBeginner]

(translated by Google translator)

Hello. 

I have published my demo program that visualizes the principle of creating an Activation Function for our proprietary SB-Neuron.

RCDS306.000-1

Copy of the program in my OneDrive:
https://1drv.ms/f/s!AnoSlTzMqlL6jNx6H4Qi...w?e=8D73FE

The program displays a section of the graph, which is saved in the file "RGB to H for SB-Neuron 04.jpg". You will also find this file in my OneDrive folder.

The section processed by the demo program corresponds to the data range:
R = 112; G = 111; B = from 0 to 255 and
R = 112; g = 112; B = from 0 to 112.

The black dots on the graph represent the training data.
The red dots correspond to the Bezier curve that describes the section of the graph.
You can see how the Bezier curve approaches the training data as you adjust the parameters.

Thus, after finishing creating all three curves, all training data of the processed training data section will be packed into 5 rows of SB-Neuron data:
- three lines with coordinates of Bezier points
- and two lines with the coordinates of two separate points.

Not bad for a start. Is it true? 

[litdev]

I guess anything that trains on part of the data and works calculating results on data that was not used for the training is the start.  Then it would be something that doesn't have a known functional ralation between input and output.  Its hard to think of a good ''simple' suggestion with:

1] Lots of available data that can be split into training and calculation data that are not the same.
2] Something where there isn't a known functional relationship between input and output.

Without these, for AI to do something valuable, then either:

1] We just reproduce the training data during testing - a simple lookup would work just as well, or:
2] It would be easier to just use the known function.

So this tends to be stuff like image recognition etc which are hard to create simple examples for - tricky!

[AbsoluteBeginner]

To LitDev,

Fortunately, there is a lot of room for discussion in this topic. 

And the criterion of truth in Nature is PRACTICE, the final result. 

Let's try, watch, learn and have fun...

Right?..